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
8
7
9
~
5
8
9
3
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
6
.
pp
5
8
7
9
-
5
8
9
3
5879
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
H
y
brid
CNBL
A
a
rchitec
ture
f
o
r ac
cura
te
ea
r
thquak
e
ma
g
nitude
foreca
sting
So
m
ia
A.
Sh
a
m
s
,
Asm
a
a
M
o
ha
m
ed,
Abeer
S.
Desuk
y
,
G
a
ber
A.
E
ls
ha
ra
wy
,
Ra
nia
Sa
l
a
h E
l
-
Sa
y
ed
D
e
p
a
r
t
me
n
t
o
f
M
a
t
h
e
m
a
t
i
c
s,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
,
A
l
-
A
z
h
a
r
U
n
i
v
e
r
si
t
y
,
C
a
i
r
o
,
Eg
y
p
t
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Oct
2
8
,
2
0
2
4
R
ev
is
ed
J
u
l 1
7
,
2
0
2
5
Acc
ep
ted
Sep
1
4
,
2
0
2
5
Earth
q
u
a
k
e
p
re
d
icti
o
n
i
n
se
ism
o
lo
g
y
is ch
a
ll
e
n
g
i
n
g
d
u
e
to
su
d
d
e
n
e
v
e
n
ts an
d
lac
k
o
f
wa
rn
in
g
s
,
re
q
u
iri
n
g
r
a
p
id
d
e
tec
ti
o
n
a
n
d
a
c
c
u
ra
te
p
a
ra
m
e
ter
e
stim
a
ti
o
n
fo
r
re
a
l
-
ti
m
e
a
p
p
li
c
a
ti
o
n
s.
T
h
is
stu
d
y
p
ro
p
o
se
d
a
n
o
v
e
l
a
u
to
m
a
ti
c
e
a
rth
q
u
a
k
e
d
e
tec
ti
o
n
m
o
d
e
l
to
e
n
h
a
n
c
e
th
e
p
r
o
c
e
ss
in
g
a
n
d
a
n
a
ly
sis
o
f
se
ism
ic
d
a
ta.
Th
e
h
y
b
rid
m
o
d
e
l
c
o
m
p
rise
s
c
o
n
v
o
lu
ti
o
n
a
l
lay
e
rs,
n
o
rm
a
li
z
a
ti
o
n
tec
h
n
iq
u
e
s,
b
i
d
irec
ti
o
n
a
l
l
o
n
g
s
h
o
rt
-
term
m
e
m
o
ry
(
Bi
-
LS
TM
)
n
e
two
rk
s,
a
n
d
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
s
,
c
o
ll
e
c
ti
v
e
ly
re
fe
rre
d
to
a
s
t
h
e
h
y
b
ri
d
c
o
n
v
o
lu
ti
o
n
a
l
–
n
o
rm
a
li
z
a
ti
o
n
–
BiL
S
TM
–
a
tt
e
n
ti
o
n
(CNBLA)
m
o
d
e
l.
T
h
e
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
a
ll
o
ws
th
e
m
o
d
e
l
to
fo
c
u
s
o
n
c
rit
ica
l
se
g
m
e
n
ts
o
f
se
ism
ic
se
q
u
e
n
c
e
s,
wh
il
e
lay
e
r
n
o
rm
a
li
z
a
ti
o
n
sta
b
i
li
z
e
s
train
in
g
b
y
n
o
rm
a
li
z
in
g
a
c
ti
v
a
ti
o
n
s,
t
h
u
s
re
d
u
c
in
g
th
e
e
ffe
c
ts
o
f
i
n
p
u
t
sc
a
le
v
a
riatio
n
s.
Th
is
d
u
a
l
a
p
p
ro
a
c
h
m
it
i
g
a
tes
t
h
e
imp
a
c
t
o
f
i
n
p
u
t
sc
a
le
v
a
ria
ti
o
n
s
a
n
d
e
n
h
a
n
c
e
s
th
e
m
o
d
e
l’s
a
b
il
i
ty
t
o
e
ffe
c
ti
v
e
ly
d
e
c
o
d
e
c
o
m
p
lex
tem
p
o
ra
l
p
a
tt
e
rn
s.
T
h
e
h
y
b
rid
CNBLA
m
o
d
e
l
o
p
ti
m
ize
s
th
e
e
x
trac
ti
o
n
a
n
d
p
ro
c
e
ss
in
g
o
f
tem
p
o
ra
l
fe
a
tu
re
s fro
m
ra
w wa
v
e
fo
rm
s rec
o
rd
e
d
a
t
sin
g
le sta
ti
o
n
s,
th
e
re
b
y
imp
ro
v
i
n
g
th
e
a
c
c
u
ra
c
y
a
n
d
e
fficie
n
c
y
o
f
se
ism
ic
m
a
g
n
it
u
d
e
e
stim
a
ti
o
n
.
Th
e
p
ro
p
o
se
d
m
o
d
e
l
is
e
v
a
lu
a
ted
u
s
in
g
tw
o
d
a
tas
e
ts:
th
e
S
TE
AD
a
n
d
USG
S
a
c
h
iev
in
g
a
m
e
a
n
sq
u
a
re
e
rr
o
r
(
M
S
E)
v
a
lu
e
s
0
.
0
5
4
a
n
d
0
.
0
8
4
3
a
n
d
a
m
e
a
n
a
b
so
lu
te
e
rro
r
(
M
AE)
0
.
1
5
a
n
d
0
.
2
5
2
6
re
sp
e
c
ti
v
e
ly
.
T
h
e
h
y
b
ri
d
CNBLA
m
o
d
e
l
o
u
t
p
e
rfo
rm
s
tw
o
b
a
se
li
n
e
m
o
d
e
ls
a
n
d
fi
v
e
sta
te
-
of
-
t
h
e
-
a
rt
a
p
p
r
o
a
c
h
e
s
in
e
a
rth
q
u
a
k
e
m
a
g
n
it
u
d
e
e
stim
a
ti
o
n
,
im
p
ro
v
in
g
se
ism
ic m
o
n
it
o
ri
n
g
a
n
d
e
a
rly
wa
rn
in
g
s
y
ste
m
s.
K
ey
w
o
r
d
s
:
Atten
tio
n
m
ec
h
an
is
m
B
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
E
ar
th
q
u
a
k
e
p
r
e
d
ictio
n
L
ay
er
n
o
r
m
aliza
tio
n
Neu
r
al
n
etwo
r
k
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
:
So
m
ia
A.
Sh
am
s
Dep
ar
tm
en
t o
f
Ma
th
em
atics,
Facu
lty
o
f
Scien
ce
,
Al
-
Az
h
ar
U
n
iv
er
s
ity
Nasr
C
ity
,
C
air
o
,
1
1
7
5
4
,
E
g
y
p
t
E
m
ail: so
m
ia.
s
h
am
s
1
0
@
az
h
ar
.
ed
u
.
eg
1.
I
NT
RO
D
UCT
I
O
N
E
ar
th
q
u
a
k
es
o
cc
u
r
wh
en
th
e
g
r
o
u
n
d
m
o
v
es
an
d
s
h
ak
es,
r
elea
s
in
g
en
er
g
y
s
to
r
e
d
in
r
o
ck
s
.
T
h
ese
n
atu
r
al
d
is
aster
s
ca
n
ca
u
s
e
s
ig
n
if
ican
t
d
am
ag
e
,
f
in
an
cial
lo
s
s
,
an
d
in
ju
r
y
,
an
d
in
s
ev
er
e
ca
s
es,
m
ass
ca
s
u
alties
[
1
]
.
T
h
ey
ca
n
ca
u
s
e
m
ass
d
ea
t
h
.
E
ar
t
h
q
u
a
k
e
ea
r
ly
war
n
i
n
g
(
E
E
W
)
h
as
em
er
g
e
d
as
a
n
ef
f
e
ctiv
e
tech
n
o
l
o
g
y
f
o
r
m
itig
atin
g
th
e
im
p
ac
t
an
d
d
am
ag
e
ca
u
s
ed
b
y
ea
r
th
q
u
a
k
es.
Nin
e
c
o
u
n
tr
ies
o
r
r
eg
i
o
n
s
cu
r
r
en
tly
h
av
e
o
p
er
atio
n
al
E
E
W
s
y
s
tem
s
;
an
o
th
er
th
ir
teen
ar
e
test
in
g
s
u
ch
s
y
s
tem
s
[
2
]
.
Fo
r
as
litt
le
as
a
f
ew
s
ec
o
n
d
s
,
E
E
W
s
y
s
tem
s
ca
n
aler
t
tar
g
et
lo
ca
ti
o
n
s
b
ef
o
r
e
th
e
ar
r
iv
al
o
f
h
ar
m
f
u
l
s
eismic
wav
es,
wh
ich
is
es
s
en
tial
f
o
r
less
en
in
g
th
e
im
p
ac
t
o
f
ea
r
th
q
u
a
k
es
[
3
]
.
T
h
e
r
eg
i
o
n
al
E
E
W
s
y
s
te
m
p
r
o
v
id
es
ea
r
ly
war
n
i
n
g
b
y
d
eter
m
in
in
g
th
e
p
r
o
p
er
ties
o
f
an
ea
r
th
q
u
ak
e,
i
n
clu
d
in
g
its
tim
e,
p
o
s
itio
n
,
a
n
d
m
a
g
n
itu
d
e.
Sin
ce
th
e
ea
r
t
h
q
u
ak
e
’
s
m
ag
n
itu
d
e
p
lay
s
a
cr
u
cial
r
o
le
in
a
war
n
in
g
s
ig
n
al,
r
ap
id
an
d
p
r
ec
is
e
m
ag
n
itu
d
e
esti
m
atio
n
is
a
cr
u
cial
f
ield
o
f
r
esear
ch
f
o
r
r
e
g
io
n
al
E
E
W
s
y
s
tem
s
[
2
]
.
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
8
7
9
-
5
8
9
3
5880
Ma
g
n
itu
d
e
i
s
a
lo
g
ar
i
th
m
i
c
i
n
d
ic
ato
r
o
f
ea
r
th
q
u
ak
e
p
o
wer
.
T
h
e
R
ich
ter
s
ca
le
,
o
f
ten
k
n
o
wn
as
th
e
lo
ca
l
m
ag
n
itu
d
e
s
ca
le
,
w
a
s
d
ev
el
o
p
ed
b
y
R
ic
h
t
er
[
4
]
an
d
h
as
b
ee
n
ex
ten
s
iv
e
ly
u
s
ed
in
s
c
ien
ti
f
i
c
s
tu
d
y
an
d
f
o
r
p
r
o
v
id
in
g
q
u
ick
p
u
b
l
ic
e
ar
th
q
u
ak
e
in
f
o
r
m
a
tio
n
.
P
eo
p
l
e
u
s
u
al
ly
f
e
el
ea
r
th
q
u
ak
e
s
g
r
ea
t
er
t
h
an
2
.
5
b
u
t
ca
n
n
o
t
f
e
el
ea
r
th
q
u
ak
e
s
le
s
s
t
h
an
2
.
5
.
E
a
r
th
q
u
ak
e
s
th
a
t
ca
u
s
e
s
ig
n
if
i
ca
n
t
d
am
ag
e
h
a
v
e
m
ag
n
i
tu
d
es
g
r
e
at
er
th
an
4
.
5
[
5
]
.
Sc
ien
ti
s
ts
s
tr
i
v
e
co
n
t
in
u
al
ly
to
im
p
r
o
v
e
m
e
th
o
d
s
f
o
r
p
r
e
d
i
ct
in
g
ea
r
th
q
u
ak
e
m
ag
n
i
tu
d
e
s
to
ac
cu
r
at
el
y
p
r
ev
en
t
s
e
r
io
u
s
c
o
n
s
e
q
u
en
ce
s
.
A
lth
o
u
g
h
it
i
s
im
p
o
s
s
ib
l
e
to
p
r
ev
en
t
ea
r
th
q
u
ak
e
s
,
t
im
e
ly
an
d
ac
cu
r
at
e
w
ar
n
i
n
g
s
ca
n
s
ig
n
if
i
ca
n
t
ly
r
e
d
u
ce
th
e
ir
d
e
s
tr
u
c
ti
v
e
im
p
a
ct
.
Va
r
io
u
s
m
eth
o
d
s
e
x
i
s
t
f
o
r
p
r
ed
i
ct
in
g
ea
r
th
q
u
ak
e
m
ag
n
i
tu
d
e
s
,
in
c
lu
d
in
g
th
e
u
s
e
o
f
s
en
s
o
r
s
,
d
ev
i
ce
s
,
m
a
g
n
e
ti
c
an
d
e
le
ct
r
i
ca
l
wav
es
,
an
d
s
e
i
s
m
i
c
in
d
i
ca
to
r
s
d
er
iv
ed
f
r
o
m
h
i
s
to
r
ic
al
d
a
ta.
A
l
th
o
u
g
h
n
o
id
e
al
m
o
d
e
l
ex
i
s
t
s
,
o
n
g
o
in
g
tr
i
al
s
en
h
an
c
e
p
r
ed
ic
tio
n
ac
cu
r
ac
y
[
6
]
.
Ar
tific
ial
in
tellig
en
ce
,
p
ar
ticu
lar
ly
n
eu
r
al
n
etwo
r
k
s
[
7
]
h
as
s
h
o
wn
g
r
ea
t
p
r
o
m
is
e
in
s
o
lv
in
g
co
m
p
lex
n
o
n
lin
ea
r
p
r
o
b
lem
s
r
elate
d
to
s
eismic
ac
tiv
ity
.
Neu
r
al
n
et
wo
r
k
s
,
a
s
u
b
s
et
o
f
m
ac
h
in
e
l
ea
r
n
in
g
,
ca
n
m
o
d
el
in
tr
icate
p
atter
n
s
an
d
r
elatio
n
s
h
ip
s
in
d
ata,
m
ak
in
g
th
em
wel
l
s
u
ited
f
o
r
ea
r
th
q
u
ak
e
m
a
g
n
it
u
d
e
p
r
ed
ictio
n
[
1
]
.
B
y
an
aly
zin
g
v
ast
am
o
u
n
ts
o
f
s
eismo
lo
g
ical
d
ata,
th
ese
m
o
d
els
ca
n
id
en
tify
s
u
b
tle
f
ea
tu
r
es
th
at
tr
ad
itio
n
al
m
eth
o
d
s
m
ay
o
v
er
l
o
o
k
.
Ma
ch
in
e
lear
n
in
g
an
d
d
ata
m
in
i
n
g
tech
n
iq
u
es
o
f
f
er
r
o
b
u
s
t
m
e
th
o
d
s
f
o
r
s
tu
d
y
in
g
s
eismic
d
ata
an
d
in
d
icato
r
s
,
m
ak
in
g
th
em
ef
f
ec
tiv
e
f
o
r
h
a
n
d
lin
g
lar
g
e
d
atasets
[
8
]
.
T
h
e
s
e
tech
n
o
lo
g
ies
h
av
e
r
ev
o
lu
tio
n
ize
d
th
e
f
ield
o
f
s
e
is
m
o
lo
g
y
,
p
r
o
v
id
in
g
n
ew
in
s
i
g
h
ts
an
d
im
p
r
o
v
in
g
t
h
e
ac
c
u
r
ac
y
o
f
ea
r
th
q
u
a
k
e
p
r
ed
ictio
n
s
.
E
n
s
u
r
in
g
th
e
q
u
a
lity
an
d
ac
cu
r
ac
y
o
f
th
e
d
ata
s
et
is
cr
itical
f
o
r
th
e
p
er
f
o
r
m
an
ce
o
f
p
r
ed
ictiv
e
m
o
d
els
[
5
]
.
Dee
p
lear
n
in
g
m
o
d
els
p
er
f
o
r
m
b
etter
wh
en
in
ter
p
r
etin
g
c
o
m
p
licated
an
d
n
o
n
lin
ea
r
in
p
u
ts
u
s
in
g
th
ese
lay
er
s
f
o
r
d
im
en
s
io
n
ali
ty
r
ed
u
ctio
n
.
Dee
p
lear
n
in
g
m
o
d
els
lik
e
g
r
a
p
h
n
e
u
r
al
n
et
wo
r
k
s
(
GNN)
[
9
]
,
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
s
[
1
0
]
,
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
L
STM
)
[
1
1
]
,
B
i
-
L
STM
[
1
2
]
,
p
r
o
v
id
e
ef
f
ec
tiv
e
ap
p
r
o
ac
h
es
f
o
r
ca
p
tu
r
i
n
g
g
eo
g
r
ap
h
ical
d
ata,
in
clu
d
i
n
g
s
tatio
n
s
an
d
th
ei
r
r
elatio
n
s
h
ip
s
.
C
h
a
k
r
ab
o
r
t
y
et
a
l.
[
1
3
]
p
r
esen
ted
a
m
u
ltit
ask
in
g
d
ee
p
lear
n
in
g
m
o
d
el
ca
lled
th
e
co
n
v
o
lu
tio
n
al
r
ec
u
r
r
e
n
t
m
o
d
el
f
o
r
ea
r
th
q
u
ak
e
id
en
tific
atio
n
an
d
m
ag
n
itu
d
e
esti
m
atio
n
(
C
R
E
I
ME
)
.
I
t
ca
n
p
er
f
o
r
m
th
e
f
o
llo
win
g
ta
s
k
s
:
i
)
id
en
tify
th
e
ea
r
th
q
u
ak
e
s
ig
n
al
f
r
o
m
b
ac
k
g
r
o
u
n
d
s
eismic
n
o
is
e
,
ii
)
ca
l
cu
late
th
e
ar
r
iv
al
tim
e
o
f
th
e
f
ir
s
t
p
r
im
ar
y
wav
e
(
P
wav
e)
,
an
d
iii
)
esti
m
ate
th
e
m
ag
n
itu
d
e
u
s
in
g
th
e
r
aw
th
r
e
e
-
co
m
p
o
n
en
t
wav
ef
o
r
m
s
f
r
o
m
a
s
in
g
le
s
tatio
n
as
th
e
m
o
d
el
in
p
u
t.
B
iases
in
p
er
f
o
r
m
an
ce
ev
al
u
atio
n
m
ay
ar
is
e
f
r
o
m
v
ar
iatio
n
s
in
p
r
ep
r
o
ce
s
s
in
g
tech
n
iq
u
es
an
d
in
p
u
t
d
ata
len
g
t
h
wh
e
n
co
m
p
ar
in
g
C
R
E
I
ME
with
o
th
er
m
o
d
els
in
th
e
s
tu
d
y
.
T
h
is
d
is
p
ar
ity
in
d
icate
s
th
at
u
n
if
o
r
m
b
en
c
h
m
ar
k
s
ar
e
n
ec
es
s
ar
y
to
g
u
a
r
an
tee
eq
u
itab
le
co
m
p
ar
is
o
n
s
ac
r
o
s
s
v
ar
io
u
s
ap
p
r
o
ac
h
es.
Saad
et
a
l.
[
1
4
]
in
tr
o
d
u
ce
d
a
m
o
d
el
co
m
p
r
is
in
g
two
s
p
ec
ial
ized
v
is
io
n
tr
a
n
s
f
o
r
m
e
r
(
ViT
)
n
etwo
r
k
s
:
o
n
e
f
o
r
id
e
n
tify
in
g
P
-
wav
e
ar
r
iv
al
tim
es
an
d
an
o
th
er
f
o
r
p
r
ed
ictin
g
ea
r
th
q
u
a
k
e
m
ag
n
itu
d
es,
b
o
th
en
g
in
ee
r
e
d
to
p
r
o
ce
s
s
s
eismic
d
ata
f
ast
a
n
d
r
eliab
ly
.
A
wid
er
r
an
g
e
o
f
p
er
f
o
r
m
a
n
ce
m
ea
s
u
r
em
en
ts
wo
u
ld
b
e
b
en
ef
icial
f
o
r
th
e
p
ap
e
r
,
ev
e
n
th
o
u
g
h
th
e
ev
alu
atio
n
m
etr
ics
em
p
lo
y
e
d
lik
e
m
ea
n
ab
s
o
lu
te
e
r
r
o
r
(
M
AE
)
ar
e
s
ig
n
if
ican
t.
Mo
u
s
av
i
an
d
B
er
o
za
[
1
5
]
u
s
e
d
co
n
v
o
lu
tio
n
al
an
d
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
,
n
am
ely
b
i
d
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
i
-
L
STM
)
u
n
it
s
,
to
ef
f
icien
tly
p
r
ed
ict
th
e
co
r
r
elatio
n
s
b
etwe
en
s
eismic
wa
v
e
am
p
litu
d
es
an
d
m
ag
n
itu
d
es.
T
h
e
tr
a
n
s
f
o
r
m
er
t
ec
h
n
iq
u
e
was
u
tili
ze
d
to
f
o
r
ec
ast
ea
r
th
q
u
ak
e
m
ag
n
it
u
d
es
b
as
ed
o
n
ex
is
tin
g
d
ata
f
o
r
th
e
Ho
r
n
o
f
A
f
r
ica
[
1
6
]
.
Sev
er
al
s
tu
d
ies
h
av
e
ex
am
in
e
d
th
e
u
s
e
o
f
v
a
r
io
u
s
m
ac
h
i
n
e
lear
n
in
g
an
d
d
ee
p
lear
n
in
g
m
o
d
els f
o
r
ea
r
th
q
u
ak
e
p
r
ed
ictio
n
t
h
at
ar
e
s
u
m
m
a
r
ized
in
T
ab
le
1
.
T
h
e
r
esu
lts
o
f
th
is
r
esear
ch
h
av
e
s
ig
n
if
ican
t
co
n
s
eq
u
en
ce
s
f
o
r
b
o
th
th
e
s
cien
tific
co
m
m
u
n
ity
an
d
p
u
b
lic
s
af
ety
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
b
u
ilt u
p
o
n
s
ix
k
e
y
co
n
tr
ib
u
tio
n
s
:
a.
I
n
teg
r
atio
n
o
f
atten
tio
n
m
ec
h
an
is
m
with
b
id
ir
ec
tio
n
al
L
STM
:
T
h
is
s
er
v
es
as
th
e
b
a
s
elin
e
m
o
d
el
f
o
r
ea
r
th
q
u
ak
e
p
r
e
d
ictio
n
,
lev
e
r
ag
in
g
th
e
s
tr
en
g
t
h
s
o
f
b
o
th
ap
p
r
o
ac
h
es.
b.
E
n
h
an
ce
d
p
r
e
d
ictio
n
th
r
o
u
g
h
lay
er
n
o
r
m
aliza
tio
n
:
B
y
r
e
p
lacin
g
th
e
atten
tio
n
m
ec
h
a
n
is
m
with
lay
er
n
o
r
m
aliza
tio
n
(
L
N)
,
th
e
s
tu
d
y
d
em
o
n
s
tr
ates
th
e
ef
f
ec
tiv
e
n
ess
o
f
th
is
ap
p
r
o
ac
h
in
m
o
d
els
th
at
d
o
n
o
t
u
tili
ze
atten
tio
n
.
c.
Dev
elo
p
m
en
t
o
f
th
e
h
y
b
r
id
co
n
v
o
l
u
tio
n
al
–
n
o
r
m
aliza
tio
n
–
B
iLST
M
–
atten
tio
n
(
C
NB
L
A)
m
o
d
el:
T
h
is
m
o
d
el
ef
f
ec
tiv
el
y
co
m
b
i
n
es
th
e
ad
v
an
ta
g
es
o
f
th
e
atten
tio
n
m
ec
h
an
is
m
an
d
lay
er
n
o
r
m
aliza
tio
n
,
wh
ich
en
h
an
ce
s
th
e
s
tab
ilit
y
o
f
t
h
e
tr
ain
in
g
p
r
o
ce
s
s
.
d.
T
h
e
cu
s
to
m
lo
s
s
f
u
n
ctio
n
is
cr
af
ted
to
allo
w
t
h
e
m
o
d
el
to
lear
n
b
o
th
ac
cu
r
ate
p
r
e
d
ictio
n
s
an
d
th
e
ass
o
ciate
d
u
n
ce
r
tain
ty
,
s
p
ec
i
f
ically
ad
d
r
ess
in
g
alea
to
r
ic
u
n
ce
r
tain
ty
,
wh
ich
r
e
f
er
s
to
th
e
u
n
ce
r
tain
ty
in
h
er
en
t in
t
h
e
d
ata
its
elf
.
e.
C
o
m
p
ar
ativ
e
an
aly
s
is
o
f
ar
ch
itectu
r
es:
T
h
e
r
esear
ch
i
n
clu
d
es
a
d
etailed
co
m
p
ar
i
s
o
n
o
f
v
a
r
io
u
s
ar
ch
itectu
r
es,
ac
co
m
p
an
ied
b
y
an
in
-
d
e
p
th
d
is
cu
s
s
io
n
o
f
th
e
r
esu
lts
o
b
tain
ed
.
f.
T
h
e
ef
f
icien
c
y
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
th
o
r
o
u
g
h
ly
e
v
alu
ated
u
s
in
g
two
d
if
f
e
r
en
t
d
atasets
an
d
s
ev
er
al
p
er
f
o
r
m
an
ce
m
et
r
ics,
in
clu
d
in
g
m
ea
n
s
q
u
ar
e
e
r
r
o
r
(
MSE
)
,
m
ea
n
ab
s
o
lu
te
e
r
r
o
r
(
MA
E
)
,
s
tan
d
ar
d
d
ev
iatio
n
o
f
m
ea
n
ab
s
o
lu
te
er
r
o
r
(
MA
E
_
STD)
,
s
tan
d
ar
d
d
ev
iatio
n
o
f
m
ea
n
s
q
u
ar
e
er
r
o
r
(
MSE
_
STD)
,
an
d
m
ea
n
co
m
b
in
atio
n
er
r
o
r
(
MCE)
.
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
Hyb
r
id
C
N
B
LA
a
r
ch
itectu
r
e
f
o
r
a
cc
u
r
a
te
ea
r
th
q
u
a
ke
ma
g
n
i
tu
d
e
fo
r
ec
a
s
tin
g
(
S
o
mia
A
.
S
h
a
ms
)
5881
I
m
p
r
o
v
ed
m
a
g
n
itu
d
e
esti
m
atio
n
ca
n
en
h
an
ce
th
e
r
eliab
ilit
y
o
f
E
E
W
s
y
s
tem
s
b
y
p
r
o
v
id
i
n
g
cr
itical
s
ec
o
n
d
s
f
o
r
in
d
iv
id
u
als
an
d
in
f
r
astru
ctu
r
es
to
tak
e
p
r
o
tectiv
e
ac
tio
n
s
.
O
u
r
f
in
d
in
g
s
will
co
n
tr
ib
u
te
t
o
o
n
g
o
in
g
ef
f
o
r
ts
to
d
ev
elo
p
m
o
r
e
r
o
b
u
s
t
an
d
r
eliab
le
ea
r
ly
ea
r
th
q
u
ak
e
war
n
in
g
s
y
s
tem
s
.
I
t
u
ltima
te
ly
aim
in
g
to
r
e
d
u
ce
th
e
im
p
ac
t o
f
t
h
ese
n
atu
r
al
d
is
aster
s
o
n
s
o
ciety
.
T
h
e
s
tr
u
ctu
r
e
o
f
th
e
p
a
p
er
i
s
o
r
g
an
ized
as
f
o
llo
ws:
T
h
e
s
ec
o
n
d
s
ec
tio
n
p
r
o
v
i
d
es
th
e
n
ec
ess
ar
y
p
r
elim
in
ar
ies,
in
clu
d
i
n
g
f
o
u
n
d
atio
n
al
co
n
ce
p
ts
a
n
d
d
e
f
in
itio
n
s
p
er
tin
en
t
to
th
is
s
tu
d
y
.
Sec
tio
n
3
p
r
esen
ts
th
e
ar
ch
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
Hy
b
r
id
C
NB
L
A
m
o
d
el
alo
n
g
with
th
e
co
n
f
ig
u
r
atio
n
o
f
th
e
b
aselin
e
m
o
d
el.
Sectio
n
4
p
r
esen
ts
th
e
r
esu
lts
an
d
d
is
cu
s
s
io
n
,
b
e
g
in
n
in
g
b
y
d
escr
ib
in
g
th
e
d
ataset
u
s
ed
an
d
ass
ess
in
g
th
e
m
o
d
el’
s
p
e
r
f
o
r
m
an
ce
.
I
n
Sect
io
n
5
,
th
e
wo
r
k
is
co
n
clu
d
ed
with
a
s
u
m
m
a
r
y
o
f
th
e
s
ig
n
if
ican
t
co
n
tr
i
b
u
tio
n
s
an
d
r
ec
o
m
m
en
d
atio
n
s
f
o
r
f
u
tu
r
e
r
esear
ch
d
ir
ec
tio
n
s
.
T
ab
le
1
.
C
o
m
p
a
r
is
o
n
o
f
ex
is
tin
g
ea
r
th
q
u
ak
e
f
o
r
ec
asti
n
g
m
o
d
els
R
e
f
.
Te
c
h
n
i
q
u
e
s
P
r
e
d
i
c
t
e
d
v
a
r
i
a
b
l
e
s
R
a
n
g
e
o
f
ma
g
n
i
t
u
d
e
Th
e
t
y
p
e
o
f
p
r
e
d
i
c
t
i
o
n
[
1
6
]
Tr
a
n
sf
o
r
mer
a
l
g
o
r
i
t
h
m
M
a
g
n
i
t
u
d
e
M
a
g
n
i
t
u
d
e
s
>
=
3.
R
e
g
r
e
ssi
o
n
[
1
7
]
LSTM
,
G
R
U
M
a
g
n
i
t
u
d
e
o
c
c
u
r
r
e
n
c
e
,
l
o
c
a
t
i
o
n
c
l
u
st
e
r
,
a
n
d
t
i
m
e
M
a
g
n
i
t
u
d
e
>
5
.
0
C
l
u
st
e
r
i
n
g
a
n
d
r
e
g
r
e
ss
i
o
n
[
1
8
]
a
t
t
e
n
t
i
o
n
-
b
a
se
d
LS
TM
t
i
m
e
,
m
a
g
n
i
t
u
d
e
,
a
n
d
l
o
c
a
t
i
o
n
mag
n
i
t
u
d
e
>
5
R
e
g
r
e
ssi
o
n
[
1
9
]
A
u
t
o
r
e
g
r
e
ssi
v
e
i
n
t
e
g
r
a
t
e
d
mo
v
i
n
g
a
v
e
r
a
g
e
(
A
R
I
M
A
)
si
n
g
u
l
a
r
s
p
e
c
t
r
u
m
a
n
a
l
y
si
s
(SSA)
M
a
g
n
i
t
u
d
e
M
a
g
n
i
t
u
d
e
>
4
r
e
g
r
e
ss
i
o
n
[
2
0
]
a
t
t
e
n
t
i
o
n
a
n
d
B
i
-
LST
M
Ea
r
t
h
q
u
a
k
e
o
r
n
o
e
a
r
t
h
q
u
a
k
e
o
c
c
u
r
r
e
n
c
e
l
o
c
a
t
i
o
n
o
c
c
u
r
r
e
n
c
e
(
r
e
g
r
e
s
si
o
n
)
M
a
g
n
i
t
u
d
e
s
b
e
t
w
e
e
n
7
a
n
d
7
.
5
C
l
a
s
si
f
i
c
a
t
i
o
n
[
2
1
]
G
N
N
w
i
t
h
b
a
t
c
h
n
o
r
ma
l
i
z
a
t
i
o
n
a
n
d
a
n
a
t
t
e
n
t
i
o
n
m
e
c
h
a
n
i
sm
d
e
p
t
h
a
n
d
m
a
g
n
i
t
u
d
e
u
n
d
e
f
i
n
e
d
r
e
g
r
e
ss
i
o
n
2.
P
RE
L
I
M
I
NAR
I
E
S
2
.
1
.
Co
nv
o
lutio
na
l
neura
l net
wo
rk
C
o
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs)
h
av
e
d
em
o
n
s
tr
ated
ef
f
icac
y
in
v
ar
io
u
s
d
o
m
ain
s
,
s
u
ch
as
im
ag
e
p
r
o
ce
s
s
in
g
,
co
n
d
itio
n
m
o
n
ito
r
in
g
,
an
d
tim
e
s
er
ies
an
aly
s
is
.
A
C
NN
is
co
n
s
t
r
u
cted
s
eq
u
en
tially
,
lay
er
in
g
th
r
ee
p
r
im
ar
y
co
m
p
o
n
en
ts
:
co
n
v
o
lu
tio
n
,
p
o
o
lin
g
,
an
d
f
u
lly
co
n
n
ec
ted
(
FC
)
lay
er
s
[
2
2
]
.
T
h
e
co
n
v
o
l
u
tio
n
lay
er
s
co
m
p
r
is
e
a
co
llectio
n
o
f
tr
ain
ab
le
k
er
n
els
th
at
ar
e
s
p
ec
if
ically
d
esig
n
ed
to
au
to
m
atica
lly
ex
tr
ac
t
lo
ca
l
f
ea
tu
r
es
f
r
o
m
th
e
in
p
u
t
m
atr
ix
[
2
3
]
.
T
h
ese
k
e
r
n
els
ex
ec
u
te
co
n
v
o
lu
tio
n
o
p
e
r
atio
n
s
b
y
u
tili
zin
g
weig
h
t
s
h
ar
in
g
an
d
lo
ca
l
co
n
n
ec
tio
n
p
r
i
n
cip
les,
r
esu
ltin
g
i
n
r
ed
u
ce
d
co
m
p
u
tatio
n
al
lo
a
d
,
d
ec
r
ea
s
ed
m
o
d
el
co
m
p
lex
ity
,
an
d
im
p
r
o
v
e
d
p
e
r
f
o
r
m
a
n
ce
.
C
NNs
h
av
e
b
ee
n
ap
p
lied
to
ea
r
th
q
u
a
k
e
p
r
e
d
ictio
n
b
y
an
aly
zin
g
s
eismic
d
ata,
s
u
ch
as
wav
ef
o
r
m
s
ig
n
als
an
d
s
p
ec
tr
o
g
r
am
s
.
C
NNs
ca
n
lear
n
g
eo
g
r
ap
h
ical
an
d
tem
p
o
r
al
p
atter
n
s
in
s
eismic
d
ata,
en
ab
lin
g
th
e
d
etec
tio
n
o
f
ea
r
th
q
u
a
k
e
p
r
ec
u
r
s
o
r
s
o
r
an
o
m
alies
in
th
e
s
ig
n
als.
C
NN
s
,
also
k
n
o
wn
as f
ea
tu
r
e
lear
n
e
r
s
,
ca
n
au
to
m
atica
lly
ex
tr
ac
t r
el
ev
an
t f
ea
tu
r
es f
r
o
m
r
aw
in
p
u
t
d
ata
[
2
4
]
.
2
.
2
.
L
o
ng
s
ho
rt
-
t
er
m
m
emo
ry
L
o
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
is
a
r
ec
u
r
r
en
t
n
eu
r
al
n
et
wo
r
k
th
at
r
etain
s
tem
p
o
r
al
co
n
n
ec
tio
n
s
b
etwe
en
in
p
u
t
item
s
d
u
r
in
g
tr
ain
in
g
.
T
h
ey
ar
e
wid
ely
u
s
ed
to
s
im
u
late
s
eq
u
en
tial
d
ata,
s
u
ch
as
ea
r
th
q
u
ak
e
s
ig
n
als
[
2
5
]
.
L
STM
u
n
its
ar
e
ef
f
ec
tiv
e
f
o
r
m
ag
n
itu
d
e
esti
m
atio
n
d
u
e
to
th
eir
g
ated
m
ec
h
an
is
m
,
wh
ich
in
clu
d
es
T
an
h
an
d
Sig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
s
,
m
ak
in
g
th
em
less
s
en
s
i
tiv
e
to
u
n
n
o
r
m
alize
d
in
p
u
t.
T
h
e
p
r
o
p
o
s
ed
L
STM
ar
c
h
itectu
r
e
i
s
illu
s
tr
ated
in
Fig
u
r
e
1
.
T
h
e
L
STM
u
n
it
co
m
p
r
is
es
a
ce
ll:
a
f
o
r
g
et
g
ate,
o
u
tp
u
t
g
ate,
an
d
in
p
u
t
g
ate.
T
h
e
ce
ll
u
n
it
is
r
esp
o
n
s
ib
le
f
o
r
s
to
r
in
g
v
alu
es
at
ea
ch
tim
e
in
ter
v
al.
T
h
e
g
ates
co
n
tr
o
l
in
f
o
r
m
atio
n
th
at
e
n
ter
s
an
d
le
av
es
th
e
r
est
o
f
th
e
u
n
it.
T
h
e
f
o
r
g
et
g
ate
(
Γ
)
in
th
e
m
em
o
r
y
b
lo
ck
s
tr
u
ctu
r
e
is
m
an
ag
ed
b
y
a
b
asic
o
n
e
-
lay
e
r
n
eu
r
al
n
etwo
r
k
.
E
q
u
ati
o
n
(
1
)
ex
p
r
ess
es
h
o
w
th
is
g
ate
o
p
er
ates.
Fo
r
g
et
g
ate(
Γ
)
d
eter
m
in
es
h
o
w
m
u
ch
o
f
th
e
s
ec
tio
n
s
h
o
u
ld
b
e
r
etai
n
ed
a
n
d
wh
at
s
h
o
u
ld
b
e
d
is
ca
r
d
e
d
.
T
h
e
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
(
σ
)
is
u
s
ed
to
i
m
p
lem
en
t it.
=
(
[
<
−
1
>
,
<
>
]
+
(
1
)
wh
er
e
(
<
>
)
is
th
e
cu
r
r
en
t
in
p
u
t,
(
<
−
1
>
)
is
th
e
p
r
e
v
io
u
s
h
id
d
en
s
ta
te,
an
d
(
an
d
)
ar
e
th
e
weig
h
t
m
atr
ix
an
d
b
ias
v
ec
to
r
,
wh
ic
h
ar
e
lear
n
ed
f
r
o
m
th
e
i
n
p
u
t
tr
ain
in
g
d
ata.
An
in
p
u
t
g
ate
is
a
u
n
it
in
wh
ich
th
e
p
r
ev
io
u
s
m
em
o
r
y
b
lo
c
k
ef
f
ec
t
f
o
r
m
s
n
ew
m
em
o
r
y
.
A
s
im
p
l
e
NN
with
an
ac
tiv
atio
n
f
u
n
ctio
n
is
ca
lled
tan
h
.
T
h
ese
o
p
er
atio
n
s
ar
e
ca
lcu
lat
ed
b
y
(
2
)
,
wh
ich
ca
lc
u
lates
th
e
ca
n
d
id
ate
ce
ll
s
tate,
a
n
d
(
3
)
,
wh
ic
h
ca
lcu
lates
h
o
w
m
u
ch
o
f
t
h
e
n
ew
ce
ll st
ate
is
r
etain
ed
u
s
in
g
th
e
s
ig
m
o
id
f
u
n
ctio
n
(
σ
)
.
̃
<
>
=
ℎ
(
[
<
−
1
>
,
<
>
]
+
)
(
2
)
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
8
7
9
-
5
8
9
3
5882
=
(
[
<
−
1
>
,
<
>
]
+
)
(
3
)
Ou
tp
u
t g
ate
(
Γ
)
: I
t c
o
n
tr
o
ls
wh
at
p
ar
ts
o
f
th
e
ce
ll st
ate
will b
e
o
u
tp
u
t,
an
d
it is
d
escr
ib
ed
b
y
(
4
)
.
=
(
[
<
−
1
>
,
<
>
]
+
)
(
4
)
T
h
e
o
u
tp
u
t g
ate
is
th
e
o
u
tp
u
t
o
f
th
e
cu
r
r
en
t L
STM
b
lo
c
k
an
d
ex
p
r
ess
ed
u
s
in
g
(
5
)
an
d
(
6
)
.
<
>
=
∗
̃
<
>
+
∗
<
−
1
>
(
5
)
<
>
=
∗
ℎ
<
>
(
6
)
W
h
er
e
(
<
>
)
ca
n
b
e
ca
lcu
lated
b
y
ap
p
ly
in
g
th
e
o
u
tp
u
t
g
ate
to
th
e
h
y
p
er
b
o
lic
tan
g
e
n
t o
f
th
e
ce
l
l state.
Fig
u
r
e
1
.
T
h
e
ar
ch
itectu
r
e
o
f
L
STM
2
.
3
.
B
idi
re
ct
io
na
l
lo
ng
-
s
ho
rt
t
er
m
m
emo
ry
T
y
p
ically
,
an
in
d
iv
id
u
al
L
ST
M
o
n
ly
f
u
n
ctio
n
s
in
th
e
f
o
r
wa
r
d
d
ir
ec
tio
n
o
f
th
e
in
f
o
r
m
ati
o
n
v
alu
e.
As
a
r
esu
lt,
th
er
e
was
o
n
ly
o
n
e
wa
y
to
d
eliv
er
th
e
in
f
o
r
m
atio
n
.
T
wo
L
STM
lay
er
s
wo
r
k
t
o
g
et
h
er
in
th
e
B
i
-
L
STM
ar
ch
itectu
r
e
[
1
2
]
,
o
n
e
lay
er
h
an
d
lin
g
f
o
r
war
d
in
f
o
r
m
atio
n
p
r
o
ce
s
s
in
g
,
an
d
th
e
s
ec
o
n
d
lay
er
h
a
n
d
lin
g
b
ac
k
war
d
e
x
ec
u
tio
n
,
as
illu
s
tr
ated
in
Fig
u
r
e
2
.
T
h
is
ar
ch
itectu
r
e
is
s
u
p
er
io
r
to
s
in
g
le
L
STM
an
d
R
NN
alg
o
r
ith
m
s
r
e
g
ar
d
in
g
ea
r
t
h
q
u
ak
e
m
ag
n
itu
d
e
p
r
ed
ictio
n
d
u
e
to
its
ab
ilit
y
t
o
u
s
e
p
r
e
v
io
u
s
an
d
s
u
b
s
eq
u
en
t
in
f
o
r
m
atio
n
.
Fig
u
r
e
2
.
B
id
ir
ec
tio
n
al
L
STM
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
Hyb
r
id
C
N
B
LA
a
r
ch
itectu
r
e
f
o
r
a
cc
u
r
a
te
ea
r
th
q
u
a
ke
ma
g
n
i
tu
d
e
fo
r
ec
a
s
tin
g
(
S
o
mia
A
.
S
h
a
ms
)
5883
2
.
4
.
At
t
ent
io
n m
ec
ha
nis
m
Atten
tio
n
m
ec
h
a
n
is
m
(
A
T
T
)
p
lay
s
a
cr
u
cial
r
o
le
in
s
eq
u
en
ce
-
to
-
s
eq
u
en
ce
m
o
d
els,
p
a
r
ticu
lar
ly
wh
e
n
d
ea
lin
g
with
lo
n
g
s
eq
u
e
n
ce
s
o
r
c
o
m
p
lex
p
atter
n
s
.
T
h
e
atte
n
tio
n
m
ec
h
an
is
m
allo
ws
t
h
e
m
o
d
el
to
f
o
cu
s
o
n
r
elev
an
t
p
ar
ts
o
f
th
e
i
n
p
u
t
s
eq
u
en
ce
,
r
eg
a
r
d
less
o
f
its
len
g
th
[
2
6
]
.
I
t
ass
ig
n
s
d
if
f
er
en
t
weig
h
ts
to
d
if
f
er
en
t
tim
e
s
tep
s
to
em
p
h
asize
im
p
o
r
tan
t
in
f
o
r
m
atio
n
.
No
te
th
at
atten
tio
n
im
p
r
o
v
es
m
o
d
el
p
e
r
f
o
r
m
an
ce
b
y
r
ed
u
cin
g
in
f
o
r
m
atio
n
lo
s
s
r
is
k
.
I
n
s
tead
o
f
r
ely
i
n
g
s
o
lely
o
n
th
e
f
i
n
al
h
id
d
en
s
tate
o
f
th
e
L
STM
,
th
is
m
o
d
el
co
n
s
id
er
s
all
h
id
d
en
s
tates in
ea
ch
d
ec
o
d
in
g
s
tep
.
3.
RE
S
E
ARCH
M
E
T
H
O
D
3
.
1
.
A
rc
hite
ct
ure
o
f
t
he
pro
po
s
ed
hy
brid
CNB
L
A
m
o
del
I
n
th
e
p
u
r
s
u
it
o
f
a
d
v
an
ci
n
g
th
e
co
m
p
u
tatio
n
al
p
r
o
ce
s
s
in
g
an
d
a
n
aly
s
is
o
f
s
eismic
d
ata,
th
is
p
a
p
er
in
tr
o
d
u
ce
s
a
h
y
b
r
id
C
NB
L
A
m
o
d
el
th
at
in
teg
r
ates
th
e
s
tr
en
g
th
s
o
f
c
o
n
v
o
lu
tio
n
al
lay
er
s
,
B
i
-
L
STM
,
an
d
A
T
T
with
in
n
o
v
ativ
e
r
eg
u
lar
izatio
n
tech
n
iq
u
es.
T
h
e
Hy
b
r
id
C
NB
L
A
m
o
d
el
am
alg
am
ates
th
e
k
ey
f
ea
tu
r
es
o
f
th
e
two
p
r
elim
in
ar
y
m
o
d
els
as
f
o
llo
ws.
T
h
e
f
ir
s
t
Mu
lti
-
C
N
N
-
Bi
-
L
STM
-
A
T
T
m
o
d
el
lev
er
ag
es
an
A
T
T
to
s
elec
tiv
ely
em
p
h
asize
s
ig
n
if
ican
t
s
eg
m
en
ts
o
f
in
p
u
t
s
eq
u
en
ce
s
,
th
er
eb
y
en
h
an
cin
g
th
e
m
o
d
el’
s
ab
ilit
y
to
h
an
d
le
co
m
p
lex
p
atter
n
s
o
v
er
lo
n
g
d
u
r
atio
n
s
.
I
n
c
o
n
tr
a
s
t,
th
e
s
ec
o
n
d
Mu
lti
-
C
NN
-
LN
-
Bi
-
L
STM
m
o
d
el
in
co
r
p
o
r
ates
lay
er
n
o
r
m
aliza
ti
o
n
(
L
N)
[
2
7
]
t
o
s
tab
ilize
th
e
t
r
ain
in
g
p
r
o
ce
s
s
an
d
m
itig
ate
t
h
e
im
p
ac
t
o
f
i
n
p
u
t
s
ca
le
v
ar
iatio
n
s
.
B
y
c
o
m
b
in
i
n
g
th
ese
a
p
p
r
o
ac
h
es,
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
C
NB
L
A
m
o
d
el
a
im
s
to
h
ar
n
ess
th
e
r
o
b
u
s
tn
ess
o
f
la
y
er
n
o
r
m
aliz
atio
n
an
d
th
e
p
r
ec
is
io
n
o
f
a
tten
tio
n
-
b
ased
m
ec
h
an
is
m
s
to
d
eliv
er
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
wh
en
an
aly
zin
g
th
r
ee
-
ch
an
n
el
s
eismo
g
r
am
d
at
a.
T
h
is
in
teg
r
atio
n
was
d
esig
n
ed
to
o
p
tim
ize
th
e
ex
tr
ac
tio
n
a
n
d
p
r
o
ce
s
s
in
g
o
f
tem
p
o
r
al
f
ea
tu
r
es,
th
e
r
eb
y
im
p
r
o
v
in
g
th
e
ac
cu
r
ac
y
a
n
d
ef
f
icien
cy
o
f
s
eismic
m
ag
n
itu
d
e
esti
m
atio
n
.
T
h
e
h
y
b
r
id
C
NB
L
A
ar
ch
itectu
r
e
is
s
h
o
wn
in
Fig
u
r
e
3
p
r
o
v
id
e
d
a
co
m
p
r
eh
e
n
s
iv
e
f
r
am
ewo
r
k
ca
p
ab
le
o
f
a
d
d
r
ess
in
g
th
e
c
h
allen
g
es
p
r
esen
ted
b
y
th
e
i
n
p
u
t
d
ata,
lead
i
n
g
to
a
m
o
r
e
ac
c
u
r
ate
a
n
d
r
eliab
le
p
r
ed
ictio
n
m
o
d
el.
Fig
u
r
e
3
.
Ar
ch
itectu
r
e
o
f
h
y
b
r
id
C
NB
L
A
m
o
d
el
f
o
r
ea
r
th
q
u
a
k
e
p
r
ed
ictio
n
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
u
s
ed
s
eismo
g
r
am
s
f
r
o
m
th
r
ee
ch
an
n
els,
ea
ch
co
v
er
in
g
3
0
s
,
to
p
r
ed
ict
ea
r
th
q
u
ak
e
m
ag
n
itu
d
es.
T
h
e
ar
ch
itectu
r
e
co
m
p
r
is
es
th
r
ee
co
n
v
o
l
u
tio
n
al
lay
er
s
with
f
ilter
co
u
n
ts
o
f
3
2
,
6
4
,
an
d
3
2
,
ea
ch
with
a
k
er
n
el
s
ize
o
f
3
.
L
ay
er
n
o
r
m
aliza
tio
n
w
as
u
s
ed
to
im
p
r
o
v
e
s
tab
ilit
y
a
n
d
ef
f
icien
cy
d
u
r
i
n
g
tr
ain
in
g
,
an
d
d
r
o
p
o
u
t
an
d
m
a
x
p
o
o
lin
g
wer
e
u
s
ed
to
d
ec
r
e
ase
s
p
atial
d
im
en
s
io
n
s
wh
ile
p
r
eser
v
in
g
ess
en
tial
f
ea
tu
r
es
.
T
h
e
o
u
tp
u
t
o
f
t
h
e
C
NN
was
u
s
ed
as
I
n
p
u
t
f
o
r
th
e
B
i
-
L
ST
M
ar
ch
itectu
r
e,
wh
ich
o
u
tp
e
r
f
o
r
m
ed
a
s
in
g
le
L
STM
in
p
r
ed
ictin
g
ea
r
th
q
u
ak
e
m
a
g
n
itu
d
es.
Atten
tio
n
m
ec
h
an
is
m
is
ad
d
ed
af
te
r
B
i
-
L
STM
to
ev
alu
ate
all
h
id
d
en
s
tates
at
ea
ch
d
ec
o
d
in
g
s
tep
.
T
h
e
m
o
d
el
in
co
r
p
o
r
ates
two
f
u
lly
co
n
n
ec
ted
De
n
s
e
lay
er
s
,
with
th
e
in
itial
lay
er
co
n
s
is
tin
g
o
f
6
4
u
n
its
an
d
L
2
r
eg
u
lar
izatio
n
t
o
m
in
im
ize
o
v
er
f
itti
n
g
.
Dr
o
p
o
u
t
r
eg
u
lar
izatio
n
is
im
p
lem
en
ted
to
en
h
an
ce
m
o
d
el
g
en
er
aliza
b
ilit
y
.
T
h
e
f
in
al
o
u
tp
u
t
is
cr
ea
ted
u
s
in
g
a
f
u
lly
c
o
n
n
ec
ted
lay
e
r
with
a
s
in
g
le
n
eu
r
o
n
,
a
n
d
a
lin
ea
r
a
ctiv
atio
n
f
u
n
ctio
n
is
u
s
ed
to
es
tim
ate
th
e
o
u
tp
u
t a
m
p
litu
d
e.
T
o
en
ab
le
th
e
p
r
o
p
o
s
ed
m
o
d
el
to
lear
n
b
o
th
ac
cu
r
ate
p
r
ed
ictio
n
s
an
d
th
eir
i
n
h
er
e
n
t
alea
to
r
ic
u
n
ce
r
tain
ty
(
d
ata
n
o
is
e)
,
we
u
s
e
a
cu
s
to
m
lo
s
s
f
u
n
ctio
n
(
7
)
.
T
h
is
lo
s
s
f
u
n
ctio
n
ac
h
iev
es th
is
b
y
co
m
b
i
n
in
g
two
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
8
7
9
-
5
8
9
3
5884
k
ey
co
m
p
o
n
en
ts
:
th
e
weig
h
te
d
s
q
u
ar
e
d
er
r
o
r
an
d
a
d
ir
ec
t
p
en
alty
o
n
th
e
p
r
e
d
icted
u
n
ce
r
t
ain
ty
.
T
h
is
m
eth
o
d
p
r
o
v
id
e
d
a
v
a
r
ian
t
o
f
t
h
e
co
m
m
o
n
MSE
lo
s
s
b
u
t
in
co
r
p
o
r
at
ed
an
ad
d
itio
n
al
s
ca
lin
g
f
ac
to
r
th
at
was
d
ep
en
d
en
t
o
n
th
e
ex
p
o
n
e
n
tial
f
u
n
cti
o
n
o
f
v
a
r
iab
le
s
i
.
T
h
e
p
ar
a
m
eter
s
i
is
ex
tr
ac
ted
f
r
o
m
t
h
e
ℎ
,
ten
s
o
r
,
wh
ic
h
r
ep
r
esen
ts
th
e
m
o
d
el's
p
r
e
d
ictio
n
s
.
T
h
e
ℎ
,
ten
s
o
r
is
ex
p
ec
ted
to
h
av
e
at
least
two
co
m
p
o
n
e
n
ts
alo
n
g
its
last
ax
is
:
th
e
p
r
ed
icte
d
v
alu
e
ℎ
,
an
d
th
e
s
ec
o
n
d
ar
y
p
ar
am
eter
s
i
.
T
h
e
s
i
p
ar
a
m
eter
is
u
s
ed
i
n
th
e
cu
s
to
m
lo
s
s
f
u
n
ctio
n
to
m
o
d
el
alea
to
r
ic
u
n
ce
r
tain
ty
,
wh
ich
r
ep
r
esen
ts
th
e
in
h
er
en
t
n
o
is
e
o
r
v
ar
ia
b
ilit
y
in
th
e
d
ata.
Sp
ec
if
ically
,
s
i
in
f
lu
en
ce
s
th
e
weig
h
tin
g
o
f
th
e
s
q
u
a
r
ed
er
r
o
r
ter
m
an
d
ad
d
s
a
r
eg
u
lar
izati
o
n
-
lik
e
ter
m
to
th
e
lo
s
s
.
T
h
e
ex
p
o
n
en
tial
tr
a
n
s
f
o
r
m
atio
n
0
.
5
.
ad
ju
s
ts
th
e
co
n
tr
ib
u
tio
n
o
f
th
e
s
q
u
ar
e
d
er
r
o
r
b
ased
o
n
th
e
v
alu
e
o
f
s
i
.
T
h
is
lo
s
s
f
u
n
ctio
n
is
p
ar
ticu
la
r
ly
in
ter
esti
n
g
b
ec
a
u
s
e
it
allo
ws
f
o
r
d
y
n
a
m
ic
ad
ju
s
tm
en
t
o
f
th
e
er
r
o
r
ter
m
b
ased
o
n
th
e
v
al
u
e
o
f
s
i
.
=
∑
(
0
.
5
.
.
(
|
,
−
ℎ
,
|
)
2
+
0
.
5
.
)
(
7
)
W
h
er
e
is
th
e
lo
s
s
o
f
elem
en
t
i
in
th
e
b
atch
,
an
d
s
i
is
a
p
ar
am
eter
th
at
m
o
d
u
lates
th
e
s
q
u
ar
ed
er
r
o
r
's
in
f
lu
en
ce
o
n
th
e
to
tal
lo
s
s
.
T
h
e
ex
p
o
n
e
n
tial
f
u
n
ctio
n
e
n
s
u
r
e
s
th
at
th
is
s
ca
lin
g
f
ac
to
r
d
ec
r
e
ases
as
s
i
in
cr
ea
s
es
,
wh
ich
ca
n
b
e
in
ter
p
r
ete
d
as r
e
d
u
cin
g
t
h
e
im
p
o
r
tan
ce
o
r
weig
h
t o
f
lar
g
e
er
r
o
r
s
wh
en
s
i
is
h
ig
h
.
W
e
u
s
ed
an
ad
ju
s
tab
le
lear
n
in
g
r
ate
tech
n
iq
u
e
in
o
u
r
im
p
lem
en
tatio
n
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
ad
ap
tiv
ely
m
o
d
if
ies
th
e
lear
n
in
g
r
ate
d
u
r
in
g
tr
ain
i
n
g
,
th
er
eb
y
en
a
b
lin
g
m
o
r
e
ef
f
ec
tiv
e
co
n
v
er
g
en
ce
.
W
e
ca
lcu
lated
th
e
lear
n
in
g
r
ate
u
s
i
n
g
(
8
)
:
=
−
1
∗
−
(
8
)
W
h
er
e
is
th
e
d
ec
ay
r
ate,
w
h
ich
in
th
is
ca
s
e
is
s
et
to
th
e
s
q
u
ar
e
r
o
o
t
o
f
0
.
1
.
T
h
e
v
al
u
e
o
f
lam
b
d
a
was
d
eter
m
in
ed
th
r
o
u
g
h
ex
ten
s
iv
e
ex
p
er
im
en
tatio
n
with
d
if
f
er
e
n
t
co
n
f
ig
u
r
atio
n
s
.
W
e
test
ed
a
v
ar
iety
o
f
lam
b
d
a
v
alu
es
to
f
in
d
th
e
o
p
tim
al
s
ettin
g
th
at
b
alan
ce
d
co
n
v
er
g
en
c
e
s
p
ee
d
an
d
s
tab
ilit
y
d
u
r
in
g
tr
ain
in
g
.
Ou
r
r
esu
lts
in
d
icate
d
th
at
a
lam
b
d
a
v
al
u
e
o
f
0
.
1
co
n
s
is
ten
tly
p
r
o
d
u
ce
d
f
av
o
r
a
b
le
r
esu
lts
ac
r
o
s
s
m
u
lt
ip
le
tr
ain
in
g
r
u
n
s
,
co
n
tr
ib
u
tin
g
to
a
m
o
r
e
s
tab
le
lear
n
in
g
p
r
o
ce
s
s
.
T
h
is
ch
o
ice
r
ef
lects
o
u
r
f
in
d
in
g
s
th
at
t
h
is
p
ar
ticu
lar
v
alu
e
allo
wed
th
e
m
o
d
el
to
c
o
n
v
er
g
e
ef
f
ec
tiv
ely
with
o
u
t
o
s
cillatio
n
s
o
r
d
iv
er
g
en
ce
.
T
h
e
p
atien
c
e
p
ar
am
eter
was
s
et
to
4
ep
o
ch
s
,
wh
ich
i
n
d
icate
s
th
at
th
e
lear
n
in
g
r
ate
d
ec
r
ea
s
ed
if
n
o
p
r
o
g
r
ess
was
d
etec
ted
d
u
r
i
n
g
th
is
tim
e
f
r
am
e.
T
h
e
ab
s
en
ce
o
f
a
co
o
ld
o
wn
p
er
i
o
d
(
co
o
ld
o
wn
=
0
)
p
er
m
itted
th
e
im
m
ed
iate
r
esu
m
p
tio
n
o
f
p
o
s
t
-
r
ed
u
ctio
n
.
T
h
is
ad
a
p
tiv
e
m
et
h
o
d
o
lo
g
y
en
h
a
n
ce
s
tr
ain
i
n
g
b
y
ac
h
iev
in
g
a
h
ar
m
o
n
io
u
s
e
q
u
ilib
r
iu
m
b
etwe
en
s
wif
t
in
itial
lear
n
in
g
an
d
m
et
icu
lo
u
s
r
ef
i
n
em
en
t
in
s
u
b
s
eq
u
en
t
p
h
ases
.
T
h
is
m
ay
r
esu
lt
in
en
h
a
n
ce
d
m
o
d
el
p
er
f
o
r
m
an
ce
a
n
d
ex
p
e
d
ited
co
n
v
er
g
en
ce
.
T
h
is
d
u
al
ap
p
r
o
ac
h
lev
er
ag
es
t
h
e
b
e
n
ef
its
o
f
b
o
t
h
n
o
r
m
aliza
tio
n
an
d
atten
tio
n
m
ec
h
an
is
m
s
to
en
s
u
r
e
r
o
b
u
s
tn
ess
an
d
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
in
s
eq
u
en
ce
p
r
o
ce
s
s
in
g
task
s
.
A
ll
p
ar
am
eter
s
ar
e
s
h
o
wn
in
T
ab
le
s
2
an
d
3
.
T
ab
le
2
.
Descr
ip
tio
n
o
f
p
r
o
p
o
s
ed
m
o
d
el
a
n
d
its
p
ar
am
ete
r
s
B
l
o
c
k
La
y
e
r
s (N
a
m
e
)
D
e
scri
p
t
i
o
n
a
n
d
p
a
r
a
me
t
e
r
v
a
l
u
e
s
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
b
l
o
c
k
c
o
n
v
1
d
1
D
c
o
n
v
l
a
y
e
r
w
i
t
h
3
2
f
i
l
t
e
r
s,
k
e
r
n
e
l
s
i
z
e
o
f
3
a
n
d
L2
r
e
g
u
l
a
r
i
z
a
t
i
o
n
=
0
.
0
0
1
l
a
y
e
r
_
n
o
r
ma
l
i
z
a
t
i
o
n
D
r
o
p
o
u
t
r
a
t
e
o
f
0
.
2
max
_
p
o
o
l
i
n
g
1
d
w
i
t
h
a
p
o
o
l
s
i
z
e
o
f
4
c
o
n
v
1
d
_
1
1
D
c
o
n
v
l
a
y
e
r
w
i
t
h
6
4
f
i
l
t
e
r
s,
k
e
r
n
e
l
s
i
z
e
o
f
3
a
n
d
L2
r
e
g
u
l
a
r
i
z
a
t
i
o
n
=
0
.
0
0
1
l
a
y
e
r
_
n
o
r
ma
l
i
z
a
t
i
o
n
_
1
D
r
o
p
o
u
t
r
a
t
e
o
f
0
.
2
max
_
p
o
o
l
i
n
g
1
d
w
i
t
h
a
p
o
o
l
s
i
z
e
o
f
4
c
o
n
v
1
d
_
2
1
D
c
o
n
v
l
a
y
e
r
w
i
t
h
3
2
f
i
l
t
e
r
s,
k
e
r
n
e
l
s
i
z
e
3
,
a
n
d
L
2
r
e
g
u
l
a
r
i
z
a
t
i
o
n
=
0
.
0
0
1
l
a
y
e
r
_
n
o
r
ma
l
i
z
a
t
i
o
n
_
2
D
r
o
p
o
u
t
r
a
t
e
o
f
0
.
2
max
_
p
o
o
l
i
n
g
1
d
w
i
t
h
a
p
o
o
l
s
i
z
e
o
f
4
S
e
q
u
e
n
c
e
l
e
a
r
n
i
n
g
Bi
-
LST
M
Th
e
LST
M
l
a
y
e
r
h
a
s
1
0
0
u
n
i
t
s
,
r
e
c
u
r
r
e
n
t
_
d
r
o
p
o
u
t
=
0
.
2
,
a
n
d
d
r
o
p
o
u
t
r
e
g
u
l
a
r
i
z
a
t
i
o
n
r
a
t
e
i
s
0
.
2
.
A
t
t
e
n
t
i
o
n
b
l
o
c
k
d
e
n
se
_
1
A
d
e
n
se
l
a
y
e
r
w
i
t
h
a
s
i
n
g
l
e
o
u
t
p
u
t
n
e
u
r
o
n
i
s a
p
p
l
i
e
d
t
o
t
h
e
i
n
p
u
t
u
s
i
n
g
t
h
e
t
a
n
h
a
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
.
A
t
t
e
n
t
i
o
n
P
r
e
d
i
c
t
i
o
n
b
l
o
c
k
d
e
n
se
_
2
D
e
n
se
l
a
y
e
r
w
i
t
h
6
4
u
n
i
t
s,
L2
r
e
g
u
l
a
r
i
z
a
t
i
o
n
=
0
.
0
0
1
,
a
n
d
t
h
e
R
e
LU
a
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
w
a
s
a
p
p
l
i
e
d
.
l
a
y
e
r
_
n
o
r
ma
l
i
z
a
t
i
o
n
_
3
D
r
o
p
o
u
t
r
a
t
e
o
f
0
.
2
d
e
n
se
_
3
D
e
n
se
l
a
y
e
r
w
i
t
h
6
4
u
n
i
t
s,
L2
r
e
g
u
l
a
r
i
z
a
t
i
o
n
=
0
.
0
0
1
,
a
n
d
R
e
LU
a
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
.
l
a
y
e
r
_
n
o
r
ma
l
i
z
a
t
i
o
n
_
4
D
r
o
p
o
u
t
r
a
t
e
o
f
0
.
2
d
e
n
se
_
4
A
d
e
n
se
l
a
y
e
r
i
s
a
p
p
l
i
e
d
w
i
t
h
a
l
i
n
e
a
r
a
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
.
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
Hyb
r
id
C
N
B
LA
a
r
ch
itectu
r
e
f
o
r
a
cc
u
r
a
te
ea
r
th
q
u
a
ke
ma
g
n
i
tu
d
e
fo
r
ec
a
s
tin
g
(
S
o
mia
A
.
S
h
a
ms
)
5885
T
ab
le
3
.
Par
am
eter
s
ettin
g
s
f
o
r
th
e
h
y
b
r
id
C
NB
L
A
m
o
d
el
P
a
r
a
me
t
e
r
V
a
l
u
e
N
o
.
o
f
t
r
a
i
n
i
n
g
sa
mp
l
e
s
5
0
7
3
4
N
o
.
o
f
v
a
l
i
d
a
t
i
o
n
sam
p
l
e
s
7
2
4
8
N
o
.
o
f
t
e
st
i
n
g
sam
p
l
e
s
1
4
4
9
5
N
o
.
o
f
e
p
o
c
h
s
2
0
0
B
a
t
c
h
si
z
e
2
5
6
L2
r
e
g
u
l
a
r
i
z
a
t
i
o
n
s
t
r
e
n
g
t
h
0
.
0
0
1
Lo
ss f
u
n
c
t
i
o
n
Th
e
c
u
st
o
m
l
o
ss fu
n
c
t
i
o
n
a
s i
n
(
7
)
o
p
t
i
m
i
z
e
r
A
d
a
m
a
l
g
o
r
i
t
h
m
Le
a
r
n
i
n
g
r
a
t
e
A
u
t
o
ma
t
i
c
a
l
l
y
a
d
j
u
s
t
s
t
h
e
l
e
a
r
n
i
n
g
r
a
t
e
d
u
r
i
n
g
t
r
a
i
n
i
n
g
,
a
s s
h
o
w
n
i
n
(
8
)
Ea
r
l
y
st
o
p
p
i
n
g
mo
n
i
t
o
r
=
v
a
l
i
d
a
t
i
o
n
l
o
ss
,
a
n
d
p
a
t
i
e
n
c
e
=
5
3
.
2
.
B
a
s
eline
m
o
del c
o
nfig
ur
a
t
io
n
3
.
2
.
1
.
M
ulti
CNN
-
Bi
-
L
S
T
M
-
AT
T
m
o
del
T
h
e
p
r
o
p
o
s
ed
n
eu
r
al
n
etwo
r
k
ar
ch
itectu
r
e
in
teg
r
ates
co
n
v
o
lu
tio
n
al
lay
er
s
,
B
i
-
L
STM
,
an
d
an
atten
tio
n
m
ec
h
an
is
m
.
Fo
llo
w
in
g
th
e
atten
tio
n
m
ec
h
a
n
is
m
,
th
e
ar
ch
itectu
r
e
in
clu
d
es
th
r
ee
f
u
lly
-
co
n
n
ec
ted
d
en
s
e
lay
er
s
to
p
r
ed
ict
th
e
o
u
tp
u
t
m
ag
n
itu
d
e.
T
h
e
atten
tio
n
m
ec
h
an
is
m
h
elp
s
th
e
m
o
d
el
f
o
cu
s
o
n
p
er
ti
n
en
t
in
p
u
t,
wh
ic
h
im
p
r
o
v
es
its
ca
p
ac
ity
to
d
ec
o
d
e
co
m
p
le
x
p
att
er
n
s
o
v
e
r
len
g
th
y
p
er
io
d
s
o
f
tim
e.
T
h
is
s
elec
tiv
e
atten
tio
n
en
h
a
n
ce
s
m
o
d
el
p
er
f
o
r
m
an
ce
,
p
ar
ticu
lar
ly
o
n
s
eq
u
en
ce
-
to
-
s
eq
u
e
n
ce
task
s
.
3
.
2
.
2
.
M
ulti
CNN
-
LN
-
Bi
-
L
S
T
M
m
o
del
T
h
is
m
o
d
el
m
o
d
if
ies
th
e
M
u
lti
-
C
NN
-
Bi
-
L
STM
-
A
T
T
m
o
d
el
b
y
s
u
b
s
titu
tin
g
th
e
atten
tio
n
m
ec
h
an
is
m
with
lay
er
n
o
r
m
aliza
tio
n
(
L
N
)
to
d
em
o
n
s
tr
ate
its
ef
f
ec
tiv
e
n
ess
in
a
m
o
d
el
with
o
u
t
atten
tio
n
.
L
N
is
ap
p
lied
af
ter
ea
ch
lear
n
ab
le
lay
er
,
s
u
ch
as
C
N
N
an
d
f
u
lly
co
n
n
ec
t
ed
lay
er
s
.
L
N
f
ac
ilit
ates
tr
ain
in
g
b
y
n
o
r
m
alizin
g
th
e
ac
tiv
atio
n
s
with
in
ea
ch
lay
er
,
th
er
eb
y
s
tab
ilizin
g
th
e
le
ar
n
in
g
p
r
o
ce
s
s
an
d
m
ak
in
g
it
less
s
en
s
i
tiv
e
to
th
e
s
ca
le
o
f
th
e
in
p
u
t
f
ea
tu
r
es;
h
o
wev
er
,
it
d
o
es
n
o
t
s
u
p
p
o
r
t
ca
p
tu
r
in
g
lo
n
g
-
ter
m
d
ep
e
n
d
e
n
cies
o
r
im
p
r
o
v
in
g
th
e
em
p
h
asis
o
n
cr
itical
in
p
u
ts
.
I
ts
p
r
im
ar
y
jo
b
in
th
is
s
itu
atio
n
is
to
im
p
r
o
v
e
s
ta
b
ilit
y
an
d
tr
ain
in
g
ef
f
ec
tiv
en
ess
;
h
o
wev
er
,
it
d
o
es
n
o
t
p
r
o
v
id
e
an
y
a
d
d
iti
o
n
al
in
p
u
t
weig
h
tin
g
ca
p
ab
i
liti
es,
as
atten
tio
n
m
ec
h
an
is
m
s
d
o
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Da
t
a
s
et
s
des
cr
iptio
n
4
.
1
.
1
.
Sta
nfo
rd
ea
r
t
hq
ua
k
e
da
t
a
s
et
(
ST
E
AD)
da
t
a
s
et
s
T
h
e
STE
AD
d
ataset
[
5
]
is
a
lar
g
e
-
s
ca
le,
g
lo
b
al
d
ataset
co
n
tain
in
g
two
class
es
o
f
wav
ef
o
r
m
s
:
Seis
m
ic
n
o
is
e
an
d
l
o
ca
l
ea
r
t
h
q
u
ak
e
wav
ef
o
r
m
s
,
wh
ic
h
ar
e
r
ec
o
r
d
ed
at
lo
ca
l
d
is
tan
ce
s
(
with
in
3
5
0
k
m
o
f
ea
r
th
q
u
ak
es).
STE
AD
in
cl
u
d
es
ap
p
r
o
x
im
ately
1
.
2
m
illi
o
n
wav
ef
o
r
m
s
r
ec
o
r
d
e
d
b
y
s
eismo
m
eter
s
lo
ca
ted
wo
r
ld
wid
e
an
d
r
esam
p
led
at
1
0
0
Hz,
with
a
d
u
r
atio
n
o
f
6
0
s
(
6
0
0
0
f
ea
t
u
r
es).
T
h
e
lo
ca
l
e
ar
th
q
u
ak
e
ca
te
g
o
r
y
co
m
p
r
is
es
ap
p
r
o
x
i
m
ately
1
,
0
5
0
,
0
0
0
th
r
ee
-
c
o
m
p
o
n
en
t
s
eismo
g
r
am
s
lin
k
ed
to
4
5
0
,
0
0
0
ea
r
th
q
u
ak
es
th
at
o
cc
u
r
r
e
d
b
etwe
en
J
an
u
a
r
y
1
9
8
4
a
n
d
Au
g
u
s
t
2
0
1
8
.
T
h
e
s
eismic
n
o
is
e
class
co
m
p
r
is
es
ap
p
r
o
x
im
ately
1
0
0
,
0
0
0
wav
ef
o
r
m
s
r
ec
o
r
d
ed
in
th
e
Un
ited
States
an
d
E
u
r
o
p
e
s
in
ce
2
0
0
0
.
W
e
r
eq
u
ir
e
s
eismic
wav
ef
o
r
m
s
f
r
o
m
co
n
tin
u
o
u
s
tim
e
s
er
ies
s
to
r
ed
in
th
e
ar
ch
iv
es
o
f
th
e
ea
r
th
q
u
ak
e
d
ata
m
an
ag
em
e
n
t
ce
n
ter
(
I
R
I
S
DM
C
)
,
wh
ich
is
a
co
llab
o
r
atio
n
o
f
m
an
y
r
esear
ch
o
r
g
a
n
izatio
n
s
.
T
h
er
e
ar
e
t
h
r
ee
ca
te
g
o
r
ies
o
f
ac
ce
s
s
s
tates
:
m
an
u
al
s
elec
tio
n
s
,
wh
ich
h
u
m
an
an
aly
s
ts
ch
o
o
s
e;
au
to
m
atic
s
e
lectio
n
s
,
wh
ich
a
r
e
d
eter
m
in
ed
b
y
a
u
to
m
atic
alg
o
r
ith
m
s
;
an
d
au
to
m
atic
p
ick
er
s
,
wh
ich
ar
e
s
elec
ted
u
s
in
g
an
AI
-
b
ased
m
o
d
el.
T
h
e
STE
AD
d
ataset
co
m
p
r
is
es sep
ar
ate
ar
r
ay
s
wit
h
th
r
ee
wav
e
f
o
r
m
s
R
ep
r
es
en
ti
n
g
t
h
r
ee
-
c
o
m
p
o
n
e
n
t
s
eis
m
o
g
r
a
m
s
.
E
ac
h
wa
v
e
f
o
r
m
h
as
6
0
0
0
c
h
a
r
a
cte
r
is
t
ics
.
T
o
p
r
e
p
a
r
e
f
o
r
tr
a
in
in
g
,
th
e
p
r
o
p
o
s
ed
m
o
d
e
l
d
o
es
n
o
t
u
s
e
a
ll
th
e
d
ata
f
r
o
m
t
h
e
ST
E
AD
d
at
ase
t.
I
t
ca
r
e
f
u
lly
s
e
lec
ts
a
s
m
a
lle
r
p
o
r
t
io
n
b
ase
d
o
n
s
p
ec
if
ic
r
u
le
s
.
T
h
ese
r
u
l
es
a
r
e
d
esi
g
n
e
d
t
o
en
s
u
r
e
d
a
ta
q
u
a
lit
y
a
n
d
r
e
le
v
a
n
c
e
t
o
ea
r
t
h
q
u
ak
es.
On
e
ex
am
p
l
e
o
f
t
h
ese
r
u
les
is
s
el
ec
t
in
g
o
n
ly
d
a
ta
i
n
p
u
ts
l
a
b
ele
d
“
t
r
a
ce
_
c
ate
g
o
r
y
”
as
“
e
a
r
t
h
q
u
ak
e
_
l
o
ca
l
”
.
T
h
e
co
l
u
m
n
s
p
_
tr
a
v
el
_
s
ec
o
n
d
s
,
s
o
u
r
c
e_
d
is
ta
n
ce
_
km
,
s
o
u
r
ce
_
m
a
g
n
it
u
d
e
,
an
d
s
o
u
r
ce
_
d
e
p
t
h
_
km
a
r
e
n
o
t
em
p
t
y
.
W
e
u
s
ed
ap
p
r
o
x
im
ately
3
0
0
,
0
0
0
ea
r
th
q
u
ak
e
wav
e
f
o
r
m
s
r
ec
o
r
d
ed
at
less
th
an
1
-
d
eg
r
ee
ep
ice
n
tr
al
d
is
tan
ce
s
.
T
h
e
en
tire
wav
ef
o
r
m
(
f
r
o
m
1
s
ec
o
n
d
b
e
f
o
r
e
P
to
e
n
d
o
f
S
co
d
a
)
was
eq
u
al
t
o
o
r
less
th
an
3
0
s
.
T
h
e
m
a
g
n
itu
d
e
d
is
tr
ib
u
tio
n
o
f
th
e
o
cc
u
r
r
en
ce
s
is
s
h
o
wn
in
Fig
u
r
e
4
.
All
w
av
ef
o
r
m
s
wer
e
b
an
d
-
p
ass
f
ilter
ed
b
etwe
en
1
.
0
to
4
0
.
0
Hz
with
a
s
ig
n
al
-
to
-
n
o
is
e
r
atio
g
r
ea
ter
th
an
2
0
d
b
.
T
o
test
th
e
im
p
ac
t
o
f
f
ac
to
r
s
s
u
ch
as
m
ag
n
itu
d
e
ty
p
e,
s
id
e
ef
f
ec
ts
,
r
eg
io
n
al
ef
f
ec
ts
,
an
d
s
ite
-
d
ep
e
n
d
en
t
lear
n
in
g
,
we
d
iv
id
ed
th
e
d
ata
in
to
s
m
aller
s
u
b
s
ets
r
an
g
in
g
f
r
o
m
6
0
K
to
1
4
0
K.
W
e
u
s
ed
7
0
%
o
f
ea
c
h
s
u
b
s
et
f
o
r
tr
ain
i
n
g
an
d
1
0
%
an
d
2
0
%
f
o
r
v
al
id
atio
n
an
d
test
in
g
r
eg
r
ess
iv
ity
.
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
8
7
9
-
5
8
9
3
5886
Fig
u
r
e
4
.
Ma
g
n
itu
d
e
d
is
tr
ib
u
tio
n
4
.
1
.
2
.
USG
S da
t
a
s
et
T
h
e
Un
ited
States
Geo
lo
g
ical
Su
r
v
ey
(
USGS)
p
r
o
v
id
es
r
ea
l
-
wo
r
ld
ea
r
th
q
u
ak
e
d
ata
o
n
h
is
to
r
ical
ea
r
th
q
u
ak
es
.
T
h
e
d
ataset
u
s
ed
f
o
r
th
is
r
esear
ch
f
r
o
m
h
ttp
s
:
//w
w
w
.
ka
g
g
le.
co
m/d
a
ta
s
ets/
w
a
r
co
d
er/ea
r
th
q
u
a
ke
-
d
a
ta
s
et
wh
ich
co
n
tain
s
ea
r
th
q
u
ak
e
d
ata
co
llected
f
r
o
m
th
e
USGS
web
s
ite
b
y
Kag
g
le.
T
h
is
d
ataset
in
clu
d
es
a
r
ec
o
r
d
o
f
titl
e,
m
ag
n
itu
d
e,
d
at
e,
tim
e,
in
ten
s
ity
,
m
ax
im
u
m
esti
m
ated
in
s
tr
u
m
en
tal
in
ten
s
ity
,
ts
u
n
am
i,
T
h
e
to
tal
n
u
m
b
er
o
f
s
eismic
s
tatio
n
s
,
T
h
e
lar
g
est
az
im
u
th
al
g
a
p
b
etw
ee
n
az
im
u
th
ally
ad
jace
n
t
s
tatio
n
s
,
d
ep
th
,
latitu
d
e,
lo
n
g
itu
d
e
a
n
d
c
o
u
n
tr
y
o
f
ev
er
y
ea
r
th
q
u
ak
e.
4
.
2
.
P
er
f
o
r
m
a
nce
ev
a
lua
t
io
n
W
e
ev
alu
ated
th
e
p
er
f
o
r
m
a
n
c
e
o
f
o
u
r
m
o
d
els
u
s
in
g
th
e
m
e
an
s
q
u
ar
e
er
r
o
r
(
MSE
)
[
2
8
]
,
MA
E
,
m
ea
n
ab
s
o
lu
te
er
r
o
r
s
tan
d
ar
d
d
ev
iati
o
n
(
MA
E
_
STD)
,
m
ea
n
s
q
u
a
r
e
er
r
o
r
s
tan
d
a
r
d
d
ev
iatio
n
(
MS
E
_
STD)
,
an
d
m
ea
n
co
m
b
in
atio
n
er
r
o
r
(
MCE)
.
L
et
n
b
e
th
e
n
u
m
b
er
o
f
s
am
p
les,
an
d
th
e
v
alu
es
1
,
2
,
3
,
…,
y
b
e
s
am
p
les
o
n
es
th
at
wer
e
o
b
s
er
v
ed
in
th
e
d
at
aset,
an
d
let
th
e
v
alu
es
̂
1
,
̂
2
,
̂
3
,
…,
̂
th
e
o
n
es
th
at
wer
e
p
r
ed
icted
b
y
th
e
m
o
d
el.
MSE
,
MA
E
_
STD
,
E
_
STD
,
an
d
(
MCE)
ca
n
b
e
ca
lcu
lated
u
s
in
g
(
9
)
-
(
1
3
)
.
=
1
∑
(
−
)
2
=
1
(
9
)
=
∑
|
−
̂
|
=
1
(
1
0
)
_
=
√
1
∑
(
|
−
̂
|
−
)
2
=
1
(
1
1
)
_
=
√
1
∑
(
(
−
)
2
−
)
2
=
1
(
1
2
)
=
∝
∗
+
(
1
−
∝
)
∗
√
(
1
3
)
wh
er
e
: α
is
a
weig
h
tin
g
f
ac
to
r
b
etwe
en
0
an
d
1
4
.
3
.
E
x
perim
ent
a
l
re
s
ults a
nd
dis
cu
s
s
io
n
All
test
s
wer
e
p
er
f
o
r
m
ed
u
s
in
g
Go
o
g
le'
s
Kag
g
le,
th
e
Ker
as
f
r
am
ewo
r
k
,
wh
ich
ca
n
o
p
er
ate
o
n
to
p
o
f
T
en
s
o
r
Flo
w,
an
d
th
e
Py
th
o
n
p
r
o
g
r
a
m
in
g
la
n
g
u
a
g
e.
All
ex
p
er
im
en
ts
wer
e
co
n
d
u
cted
o
n
an
NVI
DI
A
T
esla
K8
0
g
r
a
p
h
ical
p
r
o
ce
s
s
in
g
u
n
it
(
GPU
)
with
32
GB
R
AM
.
Fu
r
th
er
m
o
r
e
,
th
e
W
in
d
o
ws
1
0
o
p
er
atin
g
s
y
s
tem
was
u
s
ed
.
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
Hyb
r
id
C
N
B
LA
a
r
ch
itectu
r
e
f
o
r
a
cc
u
r
a
te
ea
r
th
q
u
a
ke
ma
g
n
i
tu
d
e
fo
r
ec
a
s
tin
g
(
S
o
mia
A
.
S
h
a
ms
)
5887
T
h
is
s
ec
tio
n
p
r
esen
ts
th
r
ee
a
d
ap
ted
p
r
ed
ictio
n
m
o
d
els.
T
h
e
f
ir
s
t
m
o
d
el
is
th
e
M
u
lti
C
NN
-
Bi
-
L
STM
-
AT
T
,
wh
ich
in
co
r
p
o
r
ates
a
n
at
ten
tio
n
m
ec
h
an
is
m
.
T
h
e
s
ec
o
n
d
m
o
d
el
is
th
e
Mu
lti
-
C
NN
-
LN
-
Bi
-
L
STM
,
wh
ich
u
tili
ze
s
lay
er
n
o
r
m
aliza
tio
n
.
Fin
ally
,
th
e
h
y
b
r
id
C
NB
L
A
m
o
d
el
in
teg
r
ates
th
e
s
tr
en
g
th
s
o
f
th
e
two
p
r
ev
io
u
s
ap
p
r
o
ac
h
es.
Sev
er
al
ex
p
er
im
en
ts
wer
e
ca
r
r
ied
o
u
t
to
e
x
p
lo
r
e
th
e
h
y
p
er
p
a
r
am
eter
v
alu
es
o
f
th
e
p
r
o
p
o
s
e
d
m
o
d
el.
T
h
e
p
ar
am
ete
r
v
alu
es t
h
at
ac
h
iev
ed
t
h
e
h
ig
h
est p
er
f
o
r
m
an
ce
in
th
is
s
tu
d
y
ar
e
s
u
m
m
ar
ized
in
T
ab
le
3
.
4
.
3
.
1
.
M
ulti
CNN
-
Bi
-
L
S
T
M
-
AT
T
m
o
del f
o
r
ea
rt
hq
ua
k
e
predict
io
n
I
n
th
is
s
ec
tio
n
,
we
as
s
es
s
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
Mu
lti
C
NN
-
Bi
-
L
STM
-
AT
T
m
o
d
el
f
o
r
ea
r
th
q
u
ak
e
p
r
ed
ictio
n
.
Fig
u
r
e
5
o
f
f
er
s
a
co
m
p
ellin
g
v
is
u
al
o
v
er
v
iew
o
f
th
e
m
o
d
el'
s
ef
f
ec
tiv
en
ess
,
d
em
o
n
s
tr
atin
g
its
s
u
p
er
io
r
ac
c
u
r
ac
y
,
co
n
s
is
ten
cy
,
an
d
ef
f
icien
cy
in
p
r
ed
ictin
g
e
ar
th
q
u
ak
e
m
ag
n
itu
d
es.
T
h
e
f
ig
u
r
e
f
ea
tu
r
es
a
lin
e
g
r
ap
h
th
at
d
ep
icts
th
e
tr
ain
in
g
an
d
v
alid
atio
n
lo
s
s
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
ac
r
o
s
s
s
ev
er
al
ep
o
ch
s
.
T
h
e
x
-
ax
is
r
ep
r
esen
ts
th
e
n
u
m
b
er
o
f
e
p
o
c
h
s
,
wh
ile
th
e
y
-
a
x
is
d
is
p
lay
s
th
e
co
r
r
esp
o
n
d
in
g
lo
s
s
v
alu
es.
A
lo
wer
lo
s
s
v
alu
e
in
d
icate
s
im
p
r
o
v
ed
m
o
d
el
p
e
r
f
o
r
m
a
n
ce
.
T
h
e
b
l
u
e
lin
e
illu
s
tr
ates
th
e
tr
ain
in
g
lo
s
s
,
wh
ic
h
b
eg
in
s
at
a
h
ig
h
v
alu
e
in
th
e
ea
r
ly
e
p
o
ch
s
an
d
d
ec
r
ea
s
es
r
ap
id
ly
.
As
tr
ain
in
g
p
r
o
g
r
ess
es,
th
e
lo
s
s
co
n
tin
u
es to
d
ec
lin
e,
alb
eit
at
a
s
lo
wer
r
ate.
I
t
was
f
o
u
n
d
th
at
th
e
m
o
d
el
le
ar
n
ed
ef
f
ec
tiv
ely
f
r
o
m
th
e
t
r
ain
in
g
s
et
af
ter
ap
p
r
o
x
im
ately
8
0
ep
o
ch
s
.
I
n
th
e
test
in
g
p
h
ase,
Mu
lti
C
NN
-
Bi
-
L
STM
-
AT
T
ac
h
iev
ed
MA
E
=
0
.
1
9
,
MA
S_
STD
=
0
.
2
1
,
MSE
=
0
.
0
8
3
,
MSE
_
STD
=
0
.
2
2
,
an
d
MCE
=
0
.
0
3
.
Fig
u
r
e
6
illu
s
tr
ates
th
e
r
elatio
n
s
h
ip
b
etwe
en
th
e
m
ea
s
u
r
e
d
an
d
p
r
e
d
icted
m
ag
n
itu
d
es.
T
h
e
h
o
r
izo
n
tal
a
x
is
r
ep
r
esen
ts
th
e
m
ea
s
u
r
e
d
m
ag
n
itu
d
e
d
o
cu
m
en
te
d
in
th
e
ea
r
th
q
u
ak
e
ev
en
t
ca
talo
g
m
ea
s
u
r
em
en
ts
,
wh
ile
th
e
v
er
tical
ax
is
r
ep
r
esen
ts
th
e
p
r
ed
icted
m
ag
n
itu
d
e
p
r
o
d
u
ce
d
b
y
th
e
m
o
d
el.
T
h
e
co
m
p
u
ted
lin
ea
r
r
e
g
r
ess
io
n
lin
e
is
d
is
p
lay
ed
b
y
th
e
d
iag
o
n
al
lin
e
at
th
e
ce
n
ter
o
f
th
e
p
lo
t.
T
h
u
s
,
we
ca
n
d
ed
u
ce
th
at
t
h
e
C
NN
-
Bi
-
L
STM
-
AT
T
m
o
d
el
d
e
m
o
n
s
tr
ates
a
s
atis
f
ac
to
r
y
f
it
f
o
r
th
e
d
ata
b
e
ca
u
s
e
th
e
d
ata
p
o
in
t
is
clo
s
e
to
th
e
th
eo
r
etica
l r
eg
r
e
s
s
io
n
lin
e.
Fig
u
r
e
5
.
T
r
ain
in
g
o
f
th
e
Mu
lti
-
C
NN
-
Bi
-
L
STM
-
AT
T
m
o
d
el
an
d
v
alid
atio
n
lo
s
s
o
v
er
a
n
o
.
o
f
ep
o
ch
s
Fig
u
r
e
6
.
Me
asu
r
e
d
an
d
p
r
e
d
icted
m
ag
n
itu
d
es o
f
th
e
Mu
lti C
NN
-
Bi
-
L
STM
-
AT
T
m
o
d
el
4
.
3
.
2
.
M
ulti
-
CNN
-
LN
-
Bi
-
L
S
T
M
m
o
del f
o
r
enha
nced
ea
rt
hq
ua
k
e
predict
io
n
T
h
e
n
ex
t
s
tep
in
v
esti
g
ates
t
h
e
ef
f
ec
t
o
f
i
n
co
r
p
o
r
atin
g
a
n
o
r
m
aliza
tio
n
lay
er
to
e
n
h
an
ce
th
e
p
er
f
o
r
m
an
ce
o
f
ea
r
th
q
u
ak
e
p
r
ed
ictio
n
.
Fig
u
r
e
7
d
em
o
n
s
tr
at
es
th
at
th
e
m
o
d
el
h
as
lear
n
ed
ef
f
ec
tiv
ely
f
r
o
m
th
e
tr
ain
in
g
s
et.
Ar
o
u
n
d
ep
o
c
h
3
5
,
th
e
tr
ai
n
in
g
l
o
s
s
p
latea
u
ed
,
wh
ile
th
e
v
alid
atio
n
lo
s
s
co
n
tin
u
ed
to
f
lu
ctu
ate
.
T
h
is
b
eh
a
v
io
r
in
d
icate
s
th
at
th
e
m
o
d
el
is
s
u
cc
ess
f
u
lly
lea
r
n
i
n
g
f
r
o
m
th
e
d
ata
d
u
r
in
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
.
T
h
e
Mu
lti
-
C
NN
-
LN
-
Bi
-
L
STM
m
o
d
el
ac
h
iev
ed
in
th
e
test
in
g
p
h
a
s
e,
MA
E
=
0
.
1
6
,
MA
S_
STD
=
0
.
1
8
,
MSE
=
0
.
0
5
8
,
MSE
_
STD
=
0
.
1
7
,
a
n
d
Me
an
c
o
m
b
in
atio
n
er
r
o
r
=
0
.
0
1
7
.
Fig
u
r
e
8
illu
s
tr
ates
th
e
r
elatio
n
s
h
i
p
b
etwe
en
th
e
m
ea
s
u
r
ed
an
d
p
r
ed
icted
m
ag
n
itu
d
es.
T
h
e
co
m
p
u
ted
lin
ea
r
r
eg
r
ess
io
n
lin
e
is
d
is
p
lay
ed
b
y
th
e
d
iag
o
n
al
lin
e
at
th
e
ce
n
ter
o
f
th
e
p
l
o
t.
T
h
u
s
,
w
e
ca
n
d
ed
u
ce
t
h
at
Mu
lti
-
C
NN
-
LN
-
Bi
-
L
STM
m
o
d
el
d
em
o
n
s
tr
ates
a
s
atis
f
ac
to
r
y
f
it f
o
r
th
e
d
ata
b
ec
au
s
e
th
e
d
at
a
p
o
in
t is clo
s
e
to
th
e
th
e
o
r
etic
al
r
eg
r
ess
io
n
lin
e.
4
.
3
.
3
.
H
y
brid
CNBL
A
m
o
del
f
o
r
im
pro
v
ing
ea
rt
hq
ua
k
e
p
re
dict
io
n
A
n
o
v
el
Hy
b
r
id
C
NB
L
A
m
o
d
el
is
p
r
o
p
o
s
ed
,
wh
ich
co
m
b
in
es
th
e
a
d
v
an
ta
g
eo
u
s
f
ea
tu
r
es
o
f
co
n
v
o
l
u
tio
n
al
lay
er
s
,
B
i
-
L
STM
,
an
d
AT
T
with
cr
ea
tiv
e
r
e
g
u
lar
izatio
n
tech
n
i
q
u
es.
T
h
e
co
m
b
in
atio
n
o
f
L
N
an
d
AT
T
r
e
d
u
ce
s
th
e
in
f
l
u
en
c
e
o
f
in
p
u
t
s
ca
le
f
lu
ctu
atio
n
s
,
a
llo
win
g
th
e
m
o
d
el
to
tr
ain
s
u
c
ce
s
s
f
u
lly
wh
ile
n
o
t
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
8
7
9
-
5
8
9
3
5888
n
eg
ativ
ely
in
f
lu
en
cin
g
th
e
d
a
ta.
T
h
is
co
m
b
i
n
atio
n
im
p
r
o
v
es
th
e
m
o
d
el'
s
r
o
b
u
s
tn
ess
,
al
lo
win
g
it
to
b
etter
tack
le
th
e
p
r
o
b
lem
s
p
o
s
ed
b
y
s
eismic
d
ata,
r
esu
ltin
g
in
h
ig
h
er
g
en
e
r
aliza
tio
n
to
p
r
ev
io
u
s
l
y
u
n
e
x
p
lo
r
ed
d
ata,
as
s
h
o
wn
in
th
e
f
o
llo
win
g
r
esu
lt.
T
h
e
in
clu
s
io
n
o
f
b
o
t
h
s
tr
ateg
ies
en
ab
les
m
o
r
e
ef
f
icien
t
lear
n
in
g
,
wh
er
e
L
N
s
u
p
p
o
r
ts
f
aster
co
n
v
er
g
e
n
ce
a
n
d
AT
en
s
u
r
es
th
at
th
e
m
o
d
el
lear
n
s
f
r
o
m
th
e
m
o
s
t
r
elev
an
t
in
p
u
t,
r
esu
ltin
g
i
n
f
aster
weig
h
t
m
o
d
if
icatio
n
s
d
u
r
in
g
tr
ain
i
n
g
.
Ou
r
h
y
b
r
id
C
NB
L
A
m
o
d
el
aim
s
to
lev
er
ag
e
th
e
r
esil
ien
ce
o
f
lay
er
n
o
r
m
aliza
tio
n
an
d
th
e
a
cc
u
r
ac
y
o
f
atten
tio
n
-
b
ased
p
r
o
ce
s
s
es
to
ac
h
iev
e
en
h
a
n
ce
d
p
er
f
o
r
m
an
ce
wh
e
n
an
aly
zin
g
th
r
ee
-
ch
a
n
n
el
s
eismo
g
r
am
d
ata.
T
o
e
n
h
an
ce
th
e
p
r
ec
is
io
n
an
d
ef
f
ec
tiv
en
ess
o
f
s
eismic
m
ag
n
itu
d
e
esti
m
atio
n
,
th
is
in
teg
r
atio
n
wa
s
d
ev
elo
p
e
d
to
o
p
tim
ize
th
e
e
x
tr
ac
tio
n
a
n
d
p
r
o
ce
s
s
in
g
o
f
te
m
p
o
r
al
i
n
f
o
r
m
atio
n
.
T
h
e
im
p
lem
e
n
tatio
n
o
f
th
e
h
y
b
r
id
C
NB
L
A
ar
ch
itectu
r
e
d
ep
icted
in
Fig
u
r
e
9
o
f
f
er
s
a
co
m
p
r
eh
e
n
s
iv
e
f
r
am
ewo
r
k
th
at
e
f
f
ec
tiv
ely
ta
ck
les
th
e
is
s
u
es
p
o
s
ed
b
y
th
e
in
p
u
t
d
ata,
r
esu
ltin
g
in
a
p
r
e
d
ictio
n
m
o
d
el
th
at
is
b
o
th
m
o
r
e
p
r
ec
is
e
an
d
r
eliab
l
e.
Fig
u
r
e
1
0
I
llu
s
tr
ates
th
e
r
elatio
n
s
h
ip
b
etwe
en
th
e
m
ea
s
u
r
ed
an
d
p
r
ed
icted
m
ag
n
itu
d
es.
T
h
e
co
m
p
u
ted
lin
ea
r
r
e
g
r
ess
io
n
lin
e
is
d
is
p
lay
ed
b
y
th
e
d
ia
g
o
n
al
lin
e
at
th
e
ce
n
ter
o
f
th
e
p
lo
t
.
T
h
u
s
,
we
ca
n
d
ed
u
ce
th
at
th
e
Hy
b
r
id
C
NB
L
A
m
o
d
el
ex
h
ib
its
a
s
at
is
f
ac
to
r
y
f
it
f
o
r
th
e
d
at
a
b
ec
au
s
e
th
e
d
ata
p
o
in
t
is
clo
s
e
to
th
e
th
eo
r
etica
l
r
eg
r
ess
io
n
lin
e.
I
n
th
e
test
in
g
p
h
ase,
th
e
h
y
b
r
id
p
r
o
p
o
s
ed
m
o
d
el
o
u
tp
e
r
f
o
r
m
ed
th
e
o
th
er
s
,
with
th
e
l
o
west M
AE
=
0
.
1
5
,
MA
S_
STD
=
0
.
1
7
,
MS
E
=
0
.
0
5
4
,
MSE
_
STD
=
0
.
1
5
,
an
d
MCE
=
0
.
0
1
5
.
Fig
u
r
e
7
.
Mu
lti
-
C
NN
-
LN
-
Bi
-
L
STM
's tr
ain
in
g
an
d
v
alid
atio
n
lo
s
s
o
v
er
a
n
o
.
o
f
e
p
o
ch
s
Fig
u
r
e
8
.
Me
asu
r
e
d
an
d
p
r
e
d
icted
m
ag
n
itu
d
e
esti
m
atio
n
b
y
Mu
lti
-
C
NN
-
LN
-
Bi
-
L
STM
Fig
u
r
e
9
.
Hy
b
r
id
C
NB
L
A
tr
ain
in
g
an
d
v
alid
atio
n
lo
s
s
o
v
er
a
n
o
.
o
f
e
p
o
ch
s
Fig
u
r
e
1
0
.
Me
asu
r
ed
an
d
p
r
ed
icted
m
ag
n
itu
d
e
esti
m
atio
n
o
f
Hy
b
r
i
d
C
NB
L
A
4
.
4
.
Co
m
pa
ra
t
iv
e
s
t
ud
y
a
nd
dis
cu
s
s
io
n
T
ab
le
4
p
r
esen
te
d
a
co
m
p
ar
a
tiv
e
an
aly
s
is
o
f
v
a
r
io
u
s
m
o
d
el
ar
ch
itectu
r
es
o
n
th
e
STE
A
D
d
ataset
d
u
r
in
g
test
in
g
p
h
ase.
T
h
e
co
m
p
ar
is
o
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
e
l
an
d
t
h
e
b
aselin
e
m
o
d
els
u
s
in
g
th
e
p
er
f
o
r
m
an
c
e
m
etr
ics ar
e
s
u
m
m
ar
ized
in
T
a
b
le
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.