T
E
L
KO
M
NI
K
A
,
V
ol
.
14,
N
o.
3,
S
ept
em
ber
20
16,
pp.
92
3
~
9
32
I
S
S
N
:
1
693
-
6
930
,
ac
c
r
edi
t
ed
A
b
y
D
IK
T
I,
D
e
c
r
e
e
N
o
:
58/
D
I
K
T
I
/
K
ep/
2013
D
O
I
:
10.
12928/
T
E
LK
O
M
N
I
K
A
.
v
1
4
i
3
.
2756
92
3
R
ec
ei
v
ed
A
p
r
il
2
,
2
01
6
;
R
e
v
i
s
ed
J
une
8
,
201
6
;
A
c
c
ept
ed
J
un
e
2
4
,
20
1
6
O
p
ti
mi
z
a
t
i
on of H
y
d
r
o
ge
n
-
f
ue
l
e
d E
ngi
ne
I
gni
ti
o
n Ti
mi
n
g
Based
o
n
L
-
M
N
eu
r
al
Ne
t
w
o
r
k A
lg
o
r
it
h
m
L
i
j
u
n
W
a
n
g
,
Y
u
a
n
L
i
u
,
Y
a
h
u
i
L
i
u
,
W
e
i
W
a
n
g
,
Y
a
n
a
n
Z
h
a
o
,
Z
h
e
n
z
h
o
n
g
Y
a
n
g
*
S
c
hoo
l
of
M
ec
ha
ni
c
al
E
ng
i
ne
er
i
n
g
,
N
or
t
h C
hi
na U
n
i
v
er
s
i
t
y
o
f
W
at
e
r
R
es
o
u
r
c
es
a
n
d
El
e
c
t
ri
c
Po
w
e
r,
Z
hengz
ho
u 45
00
45
,
H
en
an,
C
hi
na
*
C
or
r
es
po
ndi
ng a
ut
hor
,
e
-
m
a
i
l:
yzzh
o
@
163.
c
om
A
b
st
r
act
I
n v
i
ew
of
t
h
e i
m
pr
o
v
em
ent
m
eas
ur
es
o
f
t
he
opt
i
m
i
z
at
i
on c
o
nt
r
ol
a
l
gor
i
t
hm
f
or
t
he i
gni
t
i
on
sys
t
e
m
of
t
he hy
dr
oge
n
-
f
ue
l
ed e
ngi
ne
,
t
he L
-
M
neur
al
net
w
or
k
al
gor
i
t
hm
,
P
ow
el
l
ne
ur
al
n
et
w
or
k
a
l
gor
i
t
hm
and t
he
t
r
adi
t
i
ona
l
B
P
neur
a
l
net
w
or
k
al
gor
i
t
hm
ar
e us
ed t
o op
t
i
m
i
z
e t
he i
gn
i
t
i
on s
y
s
t
em
.
T
he r
es
ul
t
s
s
how
e
d t
ha
t
L
-
M
al
gor
i
t
hm
not
onl
y
c
an a
c
c
ur
at
el
y
pr
ed
i
c
t
t
he
hy
dr
oge
n
-
f
uel
ed
eng
i
ne
i
gn
i
t
i
o
n t
i
m
i
ng,
b
ut
al
s
o h
as
hi
g
h
pr
ec
i
s
i
on,
hi
gh
c
o
nv
er
gen
c
e
s
pee
d,
a
s
i
m
pl
e
m
odel
an
d
ot
her
out
s
t
a
ndi
ng
ad
v
ant
age
s
i
n
t
he
t
r
ai
ni
n
g
pr
oc
e
s
s
,
w
hi
c
h
c
a
n gr
eat
l
y
r
e
duc
e
t
he
w
or
k
l
oa
d of
hum
an
engi
n
e b
enc
h t
es
t
s
.
O
n
l
y
a
s
m
al
l
am
ou
nt
o
f
engi
n
e be
nc
h
t
es
t
i
s
c
ar
r
i
ed
o
ut
,
an
d t
he
obt
a
i
ne
d s
am
p
l
e d
at
a c
an be
us
ed t
o
pr
ed
i
c
t
t
h
e i
gn
i
t
i
o
n t
i
m
i
n
g
under
t
h
e w
hol
e w
or
k
i
ng c
on
di
t
i
o
ns
.
T
he m
ean
s
quar
e er
r
or
of
t
he o
pt
i
m
i
z
at
i
on r
e
s
ul
t
s
bas
e
d on L
-
M
al
gor
i
t
hm
ar
r
i
v
e
s
a
t
0.
0028
af
t
er
100
t
i
m
e
s
of
c
a
l
c
u
l
at
i
on,
t
h
e m
ax
i
m
um
v
al
u
e of
ab
s
ol
ut
e
er
r
or
ar
r
i
v
es
a
t
0.
245
4,
an
d t
h
e m
i
ni
m
um
v
al
u
e of
a
bs
o
l
ut
e
er
r
or
ar
r
i
v
es
at
0
.
00426
.
Ke
y
w
o
rd
s
:
H
y
dr
oge
n
-
f
ue
l
ed E
ngi
ne
,
L
-
M
A
l
gor
i
t
hm
,
N
e
ur
al
N
et
w
or
k
,
O
p
ti
m
i
z
a
ti
o
n
C
o
p
y
r
i
g
h
t
©
20
16 U
n
i
ver
si
t
a
s A
h
mad
D
ah
l
an
.
A
l
l
r
i
g
h
t
s r
eser
ved
.
1
.
I
n
tr
o
d
u
c
ti
o
n
R
es
ear
c
h on
h
y
dr
oge
n as
f
uel
of
i
nt
er
na
l
c
om
bus
t
i
o
n eng
i
n
e beg
an
i
n t
h
e m
i
ddl
e of
ni
n
et
ee
nt
h
c
ent
ur
y
.
I
t
f
el
l
b
ehi
nd
f
or
ab
out
100
y
e
ar
s
c
o
m
par
ed
w
i
t
h
t
he
r
es
e
ar
c
h
on
t
r
ad
i
t
i
ona
l
i
nt
er
n
al
c
om
bus
t
i
on e
ng
i
ne
.
A
l
t
ho
ugh
at
t
hi
s
s
t
a
ge,
t
he t
r
a
di
t
i
on
al
gas
o
l
i
ne
eng
i
ne
and
d
i
es
el
eng
i
ne
t
ec
h
no
l
og
i
es
bec
o
m
e
m
or
e
and
m
or
e
per
f
ec
t
and
m
at
ur
e,
but
i
t
’
s
una
bl
e
t
o
di
r
ec
t
l
y
us
e
t
he t
r
ad
i
t
i
ona
l
i
n
t
er
na
l
c
om
bus
t
i
o
n eng
i
n
e f
or
h
y
dr
og
en c
om
bus
t
i
on due t
o t
he
par
t
i
c
u
l
ar
i
t
y
of
h
y
dr
o
ge
n
as
a
f
uel
.
I
t
s
t
ec
hno
l
og
y
i
s
no
t
per
f
ec
t
and
t
h
er
e
i
s
no
c
om
m
er
c
i
al
pr
om
ot
i
on.
H
o
w
e
v
er
,
h
y
dr
o
gen
i
s
c
ons
i
der
e
d t
o b
e t
h
e m
os
t
pr
o
m
i
s
i
ng c
l
ea
n en
er
g
y
s
o
ur
c
e i
n t
hi
s
c
ent
ur
y
,
w
i
t
h
i
t
s
ad
v
an
t
ag
es
of
l
o
w
e
m
i
s
s
i
on,
r
ene
w
ab
l
e e
ner
g
y
and s
o o
n.
I
n t
h
e pr
o
c
es
s
o
f
c
o
m
bus
t
i
on,
t
he
i
gn
i
t
i
on
l
i
m
i
t
of
h
y
dr
og
en i
s
w
i
d
e,
t
h
e c
om
bus
t
i
o
n s
pee
d i
s
h
i
g
h an
d t
h
e c
al
or
i
f
i
c
v
al
ue
i
s
hi
g
h.
T
hi
s
r
es
ul
t
s
pr
e
-
i
gn
i
t
i
on,
i
nl
et
bac
k
f
i
r
e and ot
h
er
abnor
m
al
c
om
bus
t
i
on ph
en
om
enon i
n t
he
c
o
m
bus
t
i
on
pr
oc
es
s
of
i
nt
e
r
nal
c
om
bus
t
i
on
en
gi
n
e,
s
o
t
hat
t
he
e
ng
i
ne
c
an
not
o
p
er
at
e
nor
m
al
l
y
[
1]
.
H
o
w
t
o
av
oi
d
t
he
abn
o
r
m
al
c
om
bus
t
i
on
phe
nom
enon
w
i
t
h
out
r
e
duc
i
ng
t
h
e
o
ut
put
po
w
er
of
t
he
h
y
dr
o
gen
-
f
uel
ed
e
ngi
ne
i
s
t
he
r
es
e
ar
c
h
f
oc
us
of
t
he
r
el
e
v
ant
r
es
e
ar
c
h
er
s
i
n
v
ar
i
ous
c
ount
r
i
es
.
C
h
a
ngw
ei
,
et
al
.
,
ha
v
e
s
t
udi
ed
on
t
h
e ef
f
ec
t
of
hy
d
r
og
en
and m
et
han
ol
m
i
x
t
ur
e
c
o
m
bus
t
i
on
on
t
he
em
i
s
s
i
on c
har
ac
t
er
i
s
t
i
c
s
of
h
y
d
r
og
e
n
-
f
uel
ed
en
gi
n
e
[
2
]
.
J
.
M
.
G
om
es
A
nt
unes
,
et
al
.
,
ha
v
e s
t
udi
ed o
n t
h
e i
nf
l
u
enc
e of
t
he
hi
gh pr
es
s
ur
e di
r
ec
t
i
nj
ec
t
i
o
n t
ec
hno
l
og
y
o
n t
he
h
y
dr
o
ge
n
c
om
bus
t
i
on
c
h
a
r
ac
t
er
i
s
t
i
c
s
of
di
es
e
l
eng
i
ne,
a
nd
t
he
i
nf
l
u
enc
e
of
hom
ogeneo
us
c
har
ge
c
om
pr
es
s
i
on
i
g
ni
t
i
on
t
ec
h
no
l
og
y
a
nd
h
y
dr
o
gen
i
nj
ec
t
i
on
t
i
m
i
ng
and
l
en
gt
h
o
n
t
h
e
per
f
or
m
anc
e
of
t
he
h
y
dr
og
en
-
f
uel
e
d
e
ng
i
ne
[
3
]
.
X
in
g
-
hua,
F
u
-
s
h
u
i
L
iu
,
e
t
a
l.
,
ha
v
e
s
t
ud
i
e
d
t
he
ef
f
ec
t
s
o
f
h
y
dr
og
en
i
nj
ec
t
i
o
n
t
i
m
i
ng
und
er
d
i
f
f
er
ent
r
ot
at
i
o
n
s
pe
ed
and
e
qui
v
a
l
e
n
c
e
r
at
i
o
on
t
h
e
f
or
m
at
i
on of
h
y
dr
o
gen m
i
x
t
ur
e gas
an
d i
t
s
ef
f
ec
t
on t
he
pr
e
v
ent
i
o
n of
bac
k
f
i
r
eb
y
us
i
n
g
c
o
m
put
at
i
ona
l
f
l
u
i
d
d
y
nam
i
c
s
s
i
m
ul
at
i
on
[
4
]
.
T
he
ef
f
ec
t
of
E
G
R
s
y
s
t
em
on
t
he
c
o
m
bus
t
i
on
an
d
per
f
or
m
anc
e of
a c
o
m
pr
es
s
i
on i
gn
i
t
i
on h
y
dr
oge
n
-
f
uel
ed en
gi
ne
w
as
s
t
udi
ed b
y
V
i
n
od
Y
a
da
v
S
in
g
h
,
et
al
.,
[
5
]
.
H
o
w
e
v
e
r
,
t
hes
e s
t
ud
i
es
ar
e bas
e
d on a l
ar
ge n
um
ber
of
benc
h t
es
t
or
s
i
m
ul
at
i
on pr
oc
es
s
,
and o
n
l
y
a
par
t
of
t
he o
per
at
i
ng c
ond
i
t
i
ons
of
t
he h
y
dr
og
en
-
f
uel
ed en
gi
ne
c
har
ac
t
er
i
s
t
i
c
s
ar
e s
t
ud
i
e
d,
w
h
i
c
h c
an'
t
t
ak
e i
nt
o
ac
c
ount
t
h
e o
per
at
i
o
n,
em
i
s
s
i
on an
d
c
o
m
bus
t
i
on c
h
ar
ac
t
er
i
s
t
i
c
s
of
al
l
t
he
w
or
k
i
ng c
ond
i
t
i
on
s
of
t
he h
y
dr
og
en
-
f
uel
e
d e
n
gi
n
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1
6
9
3
-
6
930
T
E
L
KO
M
NI
K
A
V
o
l.
14
,
N
o
.
3,
S
ept
em
ber
2016
:
9
23
–
93
2
924
B
ec
aus
e
of
t
he
hi
g
h
c
om
bus
t
i
o
n
s
pee
d
an
d
l
o
w
i
g
ni
t
i
on
e
ner
g
y
of
h
y
dr
oge
n
,
t
he
opt
i
m
i
z
at
i
o
n
of
i
gni
t
i
on
s
y
s
t
em
i
s
c
ons
i
der
ed
t
o
b
e
t
he
m
os
t
e
f
f
ec
t
i
v
e
w
a
y
t
o
c
ont
r
ol
t
he
c
o
m
bus
t
i
on.
H
o
w
ev
er
,
b
as
ed o
n t
h
e pr
ev
i
ous
c
a
l
i
br
at
i
on i
gn
i
t
i
on
MA
P
,
i
t
i
s
of
t
en
unab
l
e
t
o c
ar
r
y
out
t
he c
on
t
r
ol
of
i
gn
i
t
i
on t
i
m
i
ng i
n al
l
c
ond
i
t
i
ons
,
w
hi
c
h r
equi
r
es
a l
ot
of
m
anpo
w
er
,
m
at
er
i
al
r
es
our
c
es
and t
i
m
e.
T
her
ef
or
e,
an
opt
i
m
i
z
at
i
on a
l
gor
i
t
hm
i
s
pr
opos
ed i
n t
hi
s
pa
p
er
t
o r
eal
i
z
e t
h
e
O
pt
i
m
al
C
ont
r
o
l
of
t
he
i
gn
i
t
i
on
s
y
s
t
em
of
t
he
h
y
dr
o
ge
n
-
f
uel
ed
en
gi
ne,
w
h
i
c
h
c
an
av
oi
d
a l
ot
o
f
m
anual
c
al
i
br
a
t
i
o
n
w
or
k
,
i
n
or
der
t
o
opt
i
m
i
z
e
t
he
i
gni
t
i
o
n
and
c
om
bus
t
i
on
p
r
oc
es
s
of
t
he
hy
d
r
og
en
-
f
uel
ed
en
gi
ne
an
d a
v
o
i
d t
he
ab
nor
m
al
c
om
bus
t
i
o
n p
hen
om
enon.
T
he
al
gor
i
t
hm
not
onl
y
r
e
al
i
z
es
t
h
e m
odel
i
n
g
of
non
l
i
ne
ar
m
appi
ng
m
odel
f
r
o
m
engi
ne
s
pee
d a
n
d l
o
a
d t
o t
he
opt
i
m
al
i
gn
i
t
i
on
ad
v
anc
e
an
gl
e,
bu
t
a
l
s
o s
ol
v
es
pr
ob
l
e
m
s
of
eas
i
l
y
f
a
l
l
i
ng
i
n
t
o t
he
l
oc
al
m
i
ni
m
u
m
,
l
o
w
c
o
n
v
er
ge
nc
e s
pe
ed
and
l
o
w
ac
c
ur
ac
y
b
y
t
he
t
r
ad
i
t
i
ona
l
B
P
ne
t
w
or
k
al
g
or
i
t
hm
.
T
he
s
i
m
ul
at
i
on r
es
ul
t
s
s
ho
w
t
h
at
t
he opt
i
m
al
c
ont
r
ol
s
c
hem
e o
f
t
he
i
gni
t
i
on ad
v
a
n
c
e angl
e of
t
he
hy
d
r
og
en
-
f
uel
ed
en
gi
n
e b
a
s
ed on
L
-
M
neur
a
l
net
w
or
k
i
s
of
hi
gh
ac
c
ur
ac
y
a
nd
hi
gh s
pe
ed.
O
nl
y
a s
m
al
l
am
ount
of
e
ng
i
ne
t
es
t
i
s
c
ar
r
i
e
d o
ut
,
and
t
h
e ob
t
ai
ne
d s
am
pl
e d
at
a
c
an b
e
us
ed
t
o
pr
edi
c
t
t
he
i
gn
i
t
i
on
t
i
m
i
ng
under
t
he
w
ho
l
e
w
or
k
i
ng
c
ond
i
t
i
ons
.
I
n
t
he
pr
oc
es
s
of
engi
n
e
oper
at
i
o
n,
t
he c
ur
r
ent
c
on
di
t
i
on p
ar
am
et
er
s
ar
e i
m
por
t
ed i
nt
o t
h
e opt
i
m
i
z
ed n
eur
al
n
et
w
or
k
c
ont
r
ol
s
y
s
t
em
,
and t
h
en t
h
e net
w
or
k
c
an out
p
ut
t
he c
ur
r
ent
opt
i
m
al
i
gn
i
t
i
on t
i
m
i
ng t
o r
eal
i
z
e t
h
e
opt
i
m
al
c
ont
r
o
l
.
A
nd
t
he
n
i
t
c
an
be
ex
t
end
ed t
o c
ont
r
ol
ot
h
er
op
er
at
i
on
par
am
et
er
s
of
t
h
e
eng
i
ne,
s
o
as
t
o
opt
i
m
i
z
e
t
he
c
ont
r
o
l
of
t
he
po
w
er
an
d
em
i
s
s
i
on
of
t
he
h
y
dr
o
gen
-
f
uel
ed
en
gi
ne
and
av
oi
d
ab
nor
m
al
c
om
bu
s
t
i
on.
2.
A
l
g
o
r
i
th
m
T
h
e
o
r
y
2
.
1
.
B
P
N
e
u
r
a
l
N
e
tw
o
r
k
A
l
g
o
r
i
th
m
BP
ne
ur
al
net
w
or
k
(
t
he
ba
c
k
-
pr
opagat
i
on
ne
ur
al
net
w
or
k
)
,
a
k
i
nd
of
no
n
f
ee
dbac
k
f
or
w
ar
d n
et
w
or
k
,
i
s
us
ed
t
o s
ol
v
e t
he
i
np
ut
/
o
ut
p
ut
non
l
i
n
ear
o
pt
i
m
i
z
at
i
o
n pr
o
bl
em
and t
h
e
i
nt
er
n
al
n
eur
ons
ar
e
pr
es
e
nt
ed
as
a
l
a
y
er
ed
ar
r
an
ge
m
ent
.
I
t
i
nc
l
ud
es
i
npu
t
l
a
y
e
r
,
hi
dde
n
l
a
ye
r
and o
ut
p
ut
l
a
y
er
,
i
n
w
h
i
c
h t
he h
i
dd
en
l
a
y
er
ne
ur
ons
c
a
n ha
v
e m
ul
t
i
p
l
e
l
a
y
er
s
,
t
h
e
w
ei
g
ht
e
d s
u
m
of
t
he out
put
of
eac
h neur
on and t
h
e i
n
put
of
t
he ne
x
t
l
a
y
er
of
neur
ons
[
6]
.
I
t
i
nc
l
ud
es
i
npu
t
l
a
y
er
,
h
i
d
den l
a
y
er
and
out
put
l
a
y
er
,
i
n w
hi
c
h t
he hi
dd
en
l
a
y
e
r
ne
ur
ons
c
an
ha
v
e
m
ul
t
i
pl
e
l
a
y
er
s
.
T
he w
e
i
ght
ed s
um
of
t
he neur
o
ns
out
p
ut
v
al
ue f
or
eac
h l
a
y
er
i
s
t
he
i
np
ut
v
al
ue of
a s
i
ng
l
e
neur
o
n
i
n
t
he
n
ex
t
l
a
y
er
.
T
he
w
or
k
pr
oc
es
s
i
s
di
v
i
ded
i
nt
o
t
r
a
i
n
i
ng
and
t
es
t
i
ng
pr
oc
es
s
,
and
t
h
e
t
r
ai
n
i
ng
pr
oc
es
s
i
s
di
v
i
d
ed
i
nt
o t
h
e f
or
w
ar
d
pr
op
agat
i
o
n pr
oc
es
s
of
t
he i
np
ut
i
nf
or
m
at
i
on and t
he
r
ev
er
s
e
pr
o
pag
at
i
on
pr
oc
e
s
s
of
t
he
er
r
or
.
I
n
t
he
t
r
ai
n
i
ng
pr
oc
es
s
,
t
he
c
o
nn
ec
t
i
on
w
ei
ght
s
bet
w
e
en e
ac
h t
w
o ne
ur
on
s
and t
hr
es
h
ol
ds
of
eac
h
neur
o
n i
n
neur
al
n
et
w
or
k
ar
e bac
k
w
ar
d
m
odi
f
i
ed b
y
t
he
er
r
or
bet
w
e
en t
h
e ou
t
put
v
al
ue of
t
he
o
ut
put
l
a
y
er
and
ac
t
ua
l
s
am
pl
e
v
a
l
u
e,
u
nt
i
l
t
he
er
r
or
s
at
i
s
f
i
es
a
des
i
r
e
d
ac
c
ur
ac
y
r
e
qui
r
em
ent
.
T
he
w
e
i
g
ht
s
an
d
t
hr
es
h
ol
ds
of
t
he
ne
t
w
or
k
ar
e f
i
x
ed af
t
er
t
r
ai
n
i
ng
,
an
d t
hen
ent
er
t
he t
es
t
i
n
g pr
oc
es
s
.
D
ur
i
ng t
he t
es
t
pr
o
c
es
s
,
onl
y
t
he
f
or
w
ar
d
pr
op
agat
i
o
n of
i
nf
or
m
at
i
on
ex
i
s
t
s
.
T
he
obj
ec
t
i
v
e f
unc
t
i
o
n
i
s
d
ef
i
ned
b
y
t
he
m
ean s
quar
e
er
r
or
(
MS
E
)
of
t
he
ac
t
u
al
o
ut
put
and
t
he
des
i
r
e
d
o
ut
p
ut
,
an
d
t
h
e
c
al
c
u
l
at
i
o
n
f
or
m
ul
a
i
s
der
i
v
ed
b
y
us
i
ng
t
he
gr
ad
i
e
nt
d
es
c
ent
m
et
hod.
T
he s
t
r
uc
t
ur
e of
a t
r
ad
i
t
i
ona
l
t
hr
ee
l
a
y
e
r
B
P
neur
a
l
net
w
or
k
m
odel
i
s
s
ho
w
n
i
n
F
i
gur
e 1.
F
i
gur
e
1.
T
he s
t
r
uc
t
ur
e
of
a B
P
ne
ur
al
net
w
or
k
…
X
2
X
1
X
n
……
H
i
d
den
l
ay
er
I
nput
l
ay
er
O
ut
put
l
ay
er
Y1
Ym
……
q
1
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
K
A
I
S
S
N
:
1
693
-
6
930
O
pt
i
m
i
z
at
i
on
of
H
y
dr
o
gen
-
f
uel
ed E
ng
i
n
e
I
gn
i
t
i
o
n T
i
mi
ng
B
a
s
ed
o
n L
-
M
N
eur
a
l
…
(
Li
j
un W
ang
)
925
G
r
adi
en
t
des
c
en
t
m
et
hod
us
ed
t
o
opt
i
m
i
z
e
t
h
e
w
e
i
ght
s
a
nd
t
hr
es
ho
l
ds
of
t
he
B
P
al
g
or
i
t
hm
i
s
t
h
e m
os
t
w
i
d
el
y
us
ed m
et
hod
i
n
t
he f
i
e
l
d
of
ar
t
i
f
i
c
i
al
n
eur
a
l
n
et
w
or
k
al
g
or
i
t
hm
and
i
t
has
bec
om
e one of
t
he i
m
por
t
ant
m
odel
s
of
neur
a
l
net
w
or
k
.
B
ut
i
t
has
s
om
e s
hor
t
c
o
m
i
ngs
,
s
uc
h
as
l
o
w
c
o
nv
er
ge
nc
e s
pee
d,
t
he
c
on
v
er
ge
nc
e s
pe
ed
i
s
r
el
a
t
ed
w
i
t
h t
h
e c
ho
i
c
e of
t
he
i
ni
t
i
a
l
v
a
l
ues
,
an
d i
t
i
s
e
as
y
t
o
f
al
l
i
nt
o
l
oc
al
m
i
ni
m
u
m
.
T
her
ef
or
e,
r
es
e
ar
c
her
s
ar
e
c
om
m
i
t
t
ed t
o
i
m
pr
ov
i
ng
t
h
e
ne
ur
al
n
et
w
o
r
k
opt
i
m
i
z
a
t
i
o
n
al
gor
i
t
hm
t
o
enhanc
e
i
t
s
ef
f
ec
t
i
v
enes
s
[
7
,
8
].
In
f
a
c
t,
as
l
ong
as
i
t
i
s
a
r
an
dom
s
ear
c
h
al
gor
i
t
hm
,
t
her
e
w
i
l
l
be
t
he
pr
ob
l
em
of
l
oc
al
ex
t
r
e
m
u
m
,
but
t
he
pos
s
i
bi
l
i
t
y
of
di
f
f
er
ent
s
i
z
e.
T
he
s
ear
c
h
s
t
r
at
eg
y
i
s
s
el
e
c
t
ed
ac
c
or
di
n
g
t
o
t
he
n
at
ur
e
of
t
he
t
ar
get
f
unc
t
i
on
i
s
a g
ood
al
t
er
nat
i
v
e.
2.
2.
L
-
M
A
l
g
o
r
i
th
m
Lev
enb
er
g
-
Mar
q
uar
dt
(
L
-
M
)
al
gor
i
t
hm
i
s
one
of
t
he o
pt
i
m
i
z
a
t
i
o
n a
l
g
or
i
t
hm
s
,
w
h
i
c
h
i
s
t
he
mo
s
t
w
i
d
el
y
us
ed n
onl
i
n
ear
l
eas
t
s
quar
es
opt
i
m
i
z
at
i
on
al
g
or
i
t
hm
[
9
]
,
w
i
t
h t
h
e ad
v
a
nt
ag
es
of
bot
h
gr
adi
ent
m
et
hod a
nd
N
e
w
t
on m
et
hod.
T
he a
l
gor
i
t
hm
i
s
a
l
s
o
der
i
v
e
d f
r
om
t
he
G
aus
s
-
N
ew
t
on
m
et
hod.
T
he bas
i
c
i
de
a of
t
he
N
e
w
t
on
m
et
hod i
s
t
o
r
e
pl
ac
e t
he
or
i
gi
n
al
ob
j
ec
t
i
v
e
f
unc
t
i
on
w
i
t
h
t
he
q
uadr
at
i
c
f
unc
t
i
on.
T
he
m
i
ni
m
u
m
poi
nt
of
t
he
or
i
g
i
nal
obj
ec
t
i
v
e
f
unc
t
i
on
i
s
r
e
pl
ac
e
d
b
y
t
he
m
i
ni
m
u
m
poi
nt
of
t
he
qua
dr
at
i
c
f
unc
t
i
o
n a
nd gr
a
du
a
l
l
y
ap
pr
oac
h
es
t
he
po
i
nt
.
I
f
t
he gen
er
al
obj
ec
t
i
v
e
f
unc
t
i
on
(
)
F
X
has
a
c
o
nt
i
n
uous
t
w
o
or
d
er
par
t
i
a
l
der
i
v
at
i
v
e,
(
)
k
X
i
s
t
he
near
poi
nt
of
t
he m
i
ni
m
u
m
poi
nt
of
(
)
F
X
,
t
hen
t
he T
a
y
l
or
ex
pa
ns
i
o
n at
t
he
poi
nt
(
)
k
X
is
:
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
1
2
TT
k
k
k
kk
k
F
X
F
X
F
X
X
X
X
X
H
X
X
X
≈
+∇
−
+
−
−
(
1)
W
he
r
e
,
(
)
(
)
(
)
(
)
2
=
k
k
ij
F
X
HX
x
x
∂
∂∂
(
)
,
=
1
,
2
...
ij
n
，
(
2)
I
s
t
he H
es
s
i
an m
at
r
i
x
f
or
t
he f
unc
t
i
on
(
)
F
X
at
t
he
po
i
nt
(
)
k
X
.
T
he abov
e T
a
y
l
or
bi
nom
i
a
l
ex
pans
i
on i
s
us
ed as
an
appr
ox
i
m
at
e s
ubs
t
i
t
ut
e f
u
nc
t
i
on
(
)
X
Φ
, th
a
t
i
s
:
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
(
)
1
2
TT
k
k
k
kk
k
X
F
X
F
X
XX
XX
H
X
XX
Φ
=
+∇
−
+
−
−
(
3)
I
f
t
he
f
unc
t
i
on
(
)
X
Φ
has
t
he
m
i
ni
m
u
m
v
al
u
e
po
i
nt
*
X
ϕ
,
t
hen
i
t
s
gr
adi
ent
v
al
ue
i
s
eq
ua
l
t
o
z
er
o,
n
am
el
y
(
)
0
X
∇
Φ
=
.
T
hen
t
he
e
quat
i
on
(
4)
i
s
g
et
.
(
)
(
)
(
)
(
)
(
)
(
)
*
kk
k
H
X
X
X
F
X
ϕ
−
=
−∇
(
4)
S
i
nc
e
t
h
e
poi
nt
*
X
ϕ
c
an
be
us
ed t
o r
e
pl
ac
e
t
h
e m
i
ni
m
u
m
poi
nt
a
ppr
ox
i
m
at
el
y
,
t
he
i
t
er
at
i
v
e f
or
m
ul
a i
s
:
(
)
(
)
(
)
(
)
(
)
(
)
1
1
kk
k
k
X
X
H
X
F
X
+
−
=
−∇
(
5)
T
he c
onv
er
ge
nc
e s
pe
ed of
t
he N
e
w
t
on m
et
hod
i
s
hi
gh,
but
i
t
has
s
t
r
i
c
t
r
eq
ui
r
em
ent
s
on
t
he pr
op
er
t
i
es
of
t
he obj
ec
t
i
v
e f
unc
t
i
on.
I
n a
ddi
t
i
o
n t
o
t
he f
unc
t
i
on m
us
t
hav
e f
i
r
s
t
and s
ec
ond
or
der
c
ont
i
n
uo
us
par
t
i
a
l
de
r
i
v
at
i
v
es
,
i
n or
d
er
t
o guar
a
nt
ee t
h
e s
t
abi
l
i
t
y
dec
l
i
n
e of
t
he
obj
ec
t
i
v
e
f
unc
t
i
on,
t
he H
es
s
i
an m
at
r
i
x
m
us
t
be pos
i
t
i
v
e
def
i
n
i
t
e
ev
er
y
w
h
er
e,
i
f
not
,
t
he N
e
w
t
on m
et
hod
w
i
l
l
f
a
il.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1
6
9
3
-
6
930
T
E
L
KO
M
NI
K
A
V
o
l.
14
,
N
o
.
3,
S
ept
em
ber
2016
:
9
23
–
93
2
926
I
n or
der
t
o
ov
er
c
om
e t
he def
i
c
i
enc
y
of
N
ew
t
on m
et
hod,
Le
v
en
ber
g a
nd M
ar
quar
dt
pr
opos
e
d t
he L
-
M opt
i
m
i
z
at
i
o
n al
gor
i
t
hm
.
I
n f
ac
t
,
t
he al
g
or
i
t
hm
i
s
a c
or
r
ec
t
i
on t
o t
he N
e
w
t
o
n
m
et
hod.
I
n
or
der
t
o
pr
ev
e
nt
t
h
e H
es
s
i
an
m
at
r
i
x
bei
ng s
i
ngu
l
ar
,
t
he
L
-
M
m
et
hod o
v
er
c
om
es
r
equi
r
em
ent
s
of
f
ul
l
r
a
nk
of
m
at
r
i
x
(
)
(
)
k
HX
of
t
he N
e
w
t
on
m
et
ho
d b
y
i
nt
r
o
duc
i
n
g a
da
m
pi
ng
par
am
et
er
k
λ
.
T
he i
t
er
at
i
v
e f
o
r
m
ul
a i
s
:
(
)
(
)
(
)
(
)
(
)
(
)
1
1
kk
k
k
k
X
X
H
X
I
F
X
λ
+
−
=
−
+∇
(
6)
W
h
er
e t
he par
am
et
er
0
k
λ
≥
and
i
t
c
ar
r
i
es
on
an
ada
pt
i
v
e
r
en
e
w
al
i
n
eac
h
i
t
er
at
i
on
pr
oc
e
s
s
,
and
I
i
s
t
he i
de
nt
i
t
y
m
at
r
i
x
.
L
-
M
al
g
or
i
t
hm
av
o
i
ds
t
he s
t
r
i
n
gent
r
e
qui
r
em
ent
s
of
H
es
s
i
an m
at
r
i
x
o
f
t
he
N
e
w
t
on
m
et
hod.
T
he
s
am
e
w
i
t
h
t
he
N
e
w
t
o
n
m
et
hod,
t
h
e
al
gor
i
t
hm
i
s
f
as
t
,
s
i
m
pl
e
and
eas
y
t
o
op
er
at
e
,
a
nd
i
t
i
s
es
pec
i
al
l
y
pr
oduc
t
i
v
e
w
hen
t
h
e
s
t
r
uc
t
ur
e
of
t
he
obj
ec
t
i
v
e
f
un
c
t
i
on
i
s
s
i
m
pl
e.
B
ec
aus
e
of
t
he c
o
nt
r
ol
pa
r
am
et
er
k
λ
, th
e
L
-
M
al
g
or
i
t
h
m
not
onl
y
pos
s
es
s
es
t
he
l
oc
a
l
s
ear
c
h
pr
oper
t
y
of
N
e
w
t
o
n m
et
hod
,
but
a
l
s
o p
os
s
es
s
es
t
he
gl
oba
l
c
on
v
er
g
enc
e
pr
op
er
t
y
of
t
he gr
adi
ent
m
et
hod.
I
t
r
et
ai
ns
t
he
ad
v
a
nt
ag
es
of
t
he t
w
o
a
l
g
or
i
t
h
m
s
,
t
he
num
ber
of
i
t
er
at
i
o
ns
i
s
s
m
al
l
,
t
he
ef
f
i
c
i
enc
y
of
t
he
net
w
or
k
t
r
a
i
ni
ng
i
s
hi
gh.
2
.
3
.
P
o
w
e
l
l
C
o
n
j
u
g
a
te
G
r
a
d
i
e
n
t M
e
th
o
d
P
o
w
e
l
l
a
l
g
or
i
t
hm
i
s
a
k
i
nd o
f
l
oc
al
s
ear
c
h m
et
hod
w
i
t
ho
ut
c
a
l
c
ul
a
t
i
n
g
der
i
v
at
i
v
e,
w
h
i
c
h
i
s
des
i
g
ned f
or
unc
ons
t
r
a
i
n
ed opt
i
m
i
z
at
i
on
pr
obl
em
.
P
o
w
e
l
l
al
g
or
i
t
hm
us
es
t
he s
uc
c
es
s
i
v
e
appr
ox
i
m
at
e c
onj
ug
at
e
d
i
r
e
c
t
i
on t
o s
e
ar
c
h f
or
t
he s
o
l
ut
i
on,
w
hi
c
h
c
an
qui
c
k
l
y
c
o
nv
er
ge
[
10
]
.
T
h
i
s
al
g
or
i
t
hm
i
s
di
v
i
ded
i
nt
o
s
e
v
er
a
l
s
t
ag
es
.
E
ac
h
s
t
ag
e
i
s
s
t
ar
t
ed
f
r
om
t
he
opt
i
m
al
p
oi
nt
of
t
he
l
as
t
s
t
age and
(
)
1
n
+
t
i
m
es
of
s
ear
c
h
es
ar
e d
one
i
n s
uc
c
es
s
i
on.
F
i
r
s
t
l
y
t
he
n
t
i
m
es
of
s
ear
c
h
es
ar
e
done
a
l
on
g
n
l
i
ne
ar
i
nd
epe
n
dent
d
i
r
ec
t
i
o
ns
,
a
nd
t
he
n
a
bes
t
di
r
ec
t
i
on
i
s
s
el
ec
t
ed
b
y
us
i
n
g
t
he
s
ear
c
h
r
es
ul
t
s
a
nd
t
he
f
or
m
er
n
di
r
ec
t
i
o
ns
ar
e
r
e
pl
ac
e
d
b
y
t
h
e
(
)
1
n
+
t
h
di
r
ec
t
i
on.
T
hen
a
ne
w
s
ear
c
h
s
t
age
i
s
s
t
ar
t
ed.
P
o
w
el
l
al
gor
i
t
hm
has
a
hi
g
h
c
onv
er
ge
nc
e
s
pee
d
i
n
dea
l
i
ng
w
i
t
h
a
c
l
as
s
of
opt
i
m
i
z
at
i
on
pr
ob
l
em
t
hat
t
he
obj
ec
t
i
v
e f
unc
t
i
o
n
w
hi
c
h
i
s
v
er
y
c
om
pl
ex
an
d i
t
s
f
unc
t
i
o
na
l
c
har
ac
t
er
i
s
t
i
c
s
i
s
not
c
l
e
ar
due t
o a
v
o
i
d
i
n
g t
he c
al
c
u
l
at
i
on of
t
he d
er
i
v
at
i
v
e t
er
m
.
B
ut
b
ec
aus
e of
it
s
n
s
ear
c
h di
r
ec
t
i
ons
i
n t
he
i
t
er
at
i
on
pr
oc
es
s
c
an
pr
ob
a
bl
y
t
ur
n
i
n
t
o
l
i
near
l
y
de
pen
dent
and
t
h
e
c
onj
ugat
e
d
i
r
ec
t
i
on c
a
n n
ot
be
f
or
m
ed,
w
h
i
c
h
l
ea
ds
t
o
t
he f
ai
l
ur
e
of
t
h
e a
l
g
or
i
t
hm
.
I
n
v
i
e
w
of
t
hi
s
s
i
t
uat
i
on
,
P
o
w
e
l
l
has
i
m
pr
ov
ed
t
he
or
i
gi
nal
a
l
gor
i
t
hm
,
and
af
t
er
eac
h
s
t
a
ge
of
t
he
ne
w
di
r
ec
t
i
on
s
ear
c
h,
i
t
i
s
nec
es
s
ar
y
t
o
c
hec
k
w
het
h
er
i
t
c
an
b
e
us
e
d
as
a
di
r
ec
t
s
ear
c
h
d
i
r
ec
t
i
on
f
or
t
he
nex
t
s
t
age of
i
t
er
at
i
o
n.
I
f
t
he di
r
e
c
t
i
on
d
oes
not
m
eet
t
he r
eq
ui
r
em
ent
s
,
i
t
i
s
nec
es
s
ar
y
t
o r
edet
er
m
i
ne
t
he opt
i
m
al
s
ol
ut
i
o
n i
n t
h
e s
our
c
e s
ear
c
h di
r
ec
t
i
on
gr
oups
t
hat
h
as
t
he m
ax
i
m
u
m
v
al
ue of
t
he
f
unc
t
i
on
dec
l
i
ne.
T
he
s
o
c
al
l
e
d
P
o
w
e
l
l
a
l
g
or
i
t
hm
i
s
ac
t
ual
l
y
t
he
P
o
w
e
l
l
c
or
r
ec
t
i
on
al
g
or
i
t
hm
.
T
he
d
et
er
m
i
nat
i
on c
on
di
t
i
o
n i
s
a
l
s
o k
now
n as
P
o
w
e
l
l
c
ond
i
t
i
on:
1)
F
i
r
s
t
l
y
,
t
h
e c
ont
r
as
t
f
unc
t
i
on
v
a
l
u
es
ar
e c
a
l
c
ul
a
t
ed
b
y
e
quat
i
o
n (
7)
,
(
8)
a
nd (
9)
.
(
)
(
)
(
)
10
2
k
kk
nn
X
XX
+
=
+
(
7)
(
)
(
)
10
k
F
F
X
=
,
(
)
(
)
2
k
n
F
F
X
=
,
(
)
(
)
31
k
n
F
F
X
+
=
(
8)
(
)
(
)
(
)
(
)
{
}
(
)
(
)
(
)
(
)
11
=
m
a
x
,
1
,
2
,
...
,
...
−−
∆
−
=
=
−
kk
k
k
i
i
mm
FX
FX
i
m
n
FX
FX
(
9)
W
h
er
e
(
)
k
m
S
i
s
t
he
di
r
ec
t
i
on
of
t
h
e m
ax
i
m
u
m
v
al
ue
of
t
he
ob
j
ec
t
i
v
e f
unc
t
i
on
dec
l
i
n
e,
a
n
d
(
)
(
)
k
m
F
X
i
s
t
he
obj
ec
t
i
v
e f
unc
t
i
on
v
a
l
ue.
2)
I
f
t
he f
ol
l
o
w
i
ng
i
ne
qua
l
i
t
i
es
(
)
(
)
(
)
31
2
1
2
31
2
1
3
1
2
2
FF
F
F
F
F
F
F
F
<
−
+
−
−∆
<
∆
−
(
10)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
K
A
I
S
S
N
:
1
693
-
6
930
O
pt
i
m
i
z
at
i
on
of
H
y
dr
o
gen
-
f
uel
ed E
ng
i
n
e
I
gn
i
t
i
o
n T
i
mi
ng
B
a
s
ed
o
n L
-
M
N
eur
a
l
…
(
Li
j
un W
ang
)
927
A
r
e bot
h s
at
i
s
f
i
ed,
t
he
n t
h
e di
r
ec
t
i
on n
eeds
t
o c
h
an
ge,
ot
her
w
i
s
e t
he or
i
gi
na
l
di
r
ec
t
i
on i
s
s
t
i
l
l
adop
t
ed.
3.
M
o
d
e
l
i
n
g
a
n
d
T
r
a
i
n
i
n
g
o
f L
-
M
N
e
u
r
a
l
N
e
tw
o
r
k
C
o
n
tr
o
l
S
yst
em
I
n
v
i
e
w
of
t
he
ob
v
i
ous
d
i
s
a
dv
a
nt
a
ges
of
B
P
al
g
or
i
t
hm
,
t
he
L
-
M
al
gor
i
t
hm
i
s
adopt
ed
t
o
av
o
i
d
t
h
e
l
oc
a
l
opt
i
m
al
s
ol
ut
i
o
n
i
n
t
h
e
s
ear
c
h
pr
oc
es
s
.
I
f
i
t
i
s
appl
i
ed
t
o
t
h
e
t
r
ai
ni
n
g
pr
oc
es
s
of
neur
a
l
net
w
or
k
,
i
t
c
an
j
um
p
out
of
l
oc
a
l
opt
i
m
al
s
ol
ut
i
o
ns
and
o
bt
a
i
n
t
he
g
l
ob
al
op
t
i
m
al
s
ol
ut
i
on.
T
hi
s
i
s
benef
i
c
i
a
l
t
o
i
m
pr
ov
e
t
he
r
e
l
i
ab
i
l
i
t
y
of
t
he
s
o
l
ut
i
on,
r
ed
uc
e
t
h
e
num
ber
of
i
t
er
at
i
ons
,
a
nd
i
m
pr
ov
e
t
h
e
c
on
v
er
g
enc
e
p
r
ec
i
s
i
on.
A
s
t
he
s
e
l
ec
t
e
d
o
bj
ec
t
i
v
e
f
unc
t
i
o
n
i
t
s
el
f
i
s
no
t
c
om
pl
i
c
at
ed,
t
he a
dop
t
i
o
n of
L
-
M a
l
g
or
i
t
hm
ac
hi
ev
es
a
v
er
y
g
ood
ef
f
ec
t
.
S
o i
n
t
hi
s
p
aper
,
L
-
M al
gor
i
t
hm
i
s
adop
t
ed t
o t
r
ai
n t
he n
eur
al
ne
t
w
or
k
w
hi
c
h
i
s
t
o us
e L
-
M al
gor
i
t
hm
t
o c
al
c
ul
at
e t
h
e opt
i
m
al
w
ei
g
ht
s
and t
hr
es
hol
ds
of
t
he neur
a
l
net
w
or
k
t
o get
t
he ex
ac
t
v
a
l
ue of
t
he
gl
o
bal
o
pt
i
m
al
s
ol
ut
i
on.
T
he a
l
g
or
i
t
hm
m
ak
es
t
he neur
al
net
w
or
k
m
o
del
i
ng
,
t
r
ai
ni
ng,
s
i
m
ul
at
i
on
and s
o o
n t
o
ac
hi
e
v
e t
he d
es
i
r
ed
ef
f
ec
t
s
,
s
o as
t
o ef
f
ec
t
i
v
el
y
us
e
t
he a
l
gor
i
t
hm
f
or
h
y
dr
o
gen
-
f
uel
ed en
gi
ne
i
gn
i
t
i
on
s
y
s
t
em
opt
i
m
i
z
at
i
o
n m
odel
i
ng
an
d pr
e
di
c
t
i
o
n of
t
h
e
i
gn
i
t
i
on
M
A
P
un
der
t
he
w
h
ol
e
w
or
k
i
ng c
ondi
t
i
o
ns
.
A
n
d c
o
m
par
ed w
i
t
h t
h
e t
r
ad
i
t
i
ona
l
B
P
a
l
g
or
i
t
hm
and P
o
w
e
l
l
al
g
or
i
t
hm
,
t
he
r
es
ul
t
s
s
ho
w
t
h
at
t
h
e L
-
M a
l
gor
i
t
hm
has
hi
gh
er
pr
ec
i
s
i
on,
h
i
gh
er
c
onv
er
ge
nc
e s
p
eed a
nd a
v
er
y
s
i
m
pl
e
m
odel
.
T
he
ne
ur
al
n
et
w
or
k
t
r
ai
ned
b
y
t
hi
s
a
l
gor
i
t
hm
c
an
be
us
ed
t
o
s
i
m
ul
at
e
t
he
i
g
ni
t
i
on
s
y
s
t
em
of
t
he h
y
dr
og
en
-
f
ue
l
ed
en
gi
ne.
W
i
t
h t
he c
har
ac
t
er
i
s
t
i
c
s
of
f
as
t
,
ac
c
ur
at
e
a
nd s
e
ns
i
t
i
v
e
,
i
t
c
an al
s
o
opt
i
m
i
z
e a
nd c
o
nt
r
ol
t
he h
y
dr
o
ge
n
-
f
uel
ed
en
gi
ne.
T
hr
ee
l
a
y
er
neur
a
l
n
et
w
or
k
i
s
s
el
ec
t
e
d,
t
h
e
i
n
put
s
of
t
he
i
np
ut
l
a
y
er
ar
e
t
he
h
y
dr
ogen
-
f
uel
ed
eng
i
n
e
w
or
k
i
ng
c
o
n
di
t
i
on
par
am
et
er
s
,
nam
el
y
,
t
he
e
ng
i
ne
s
pee
d
(
)
/
m
i
n
nr
and
t
h
e
l
oa
d
(
)
%
ρ
par
am
et
er
s
,
t
he
ou
t
put
of
t
he
net
w
or
k
i
s
t
he
opt
i
m
al
i
gni
t
i
o
n
a
dv
anc
e
ang
l
e
(
)
C
A
θ
°
.
So
t
he
i
n
put
l
a
y
er
of
neur
al
n
e
t
w
or
k
c
ont
ai
ns
2
neur
o
ns
,
and
t
h
e
out
put
l
a
y
er
c
o
nt
a
i
ns
1
neur
ons
.
T
he num
ber
of
neur
ons
i
n
t
he
hi
dde
n
l
a
y
er
i
s
det
er
m
i
ned b
y
ex
per
i
enc
e
or
f
or
m
ul
a.
I
n t
hi
s
m
odel
,
t
he
n
um
ber
of
neur
ons
i
n t
he
hi
dde
n
l
a
y
er
i
s
5
.
T
he
“
s
i
gm
oi
d”
f
unc
t
i
o
n
i
s
s
el
ec
t
ed
as
t
he
t
r
ans
f
er
f
unc
t
i
on
bet
w
ee
n
e
ac
h
t
w
o
neur
o
ns
.
A
f
t
er
t
he
neur
a
l
net
w
or
k
m
odel
i
s
es
t
abl
i
s
he
d,
t
he
t
r
ai
n
i
ng
pr
oc
es
s
of
t
h
e
n
et
w
or
k
c
an
be
c
ar
r
i
ed
out
.
I
n
t
hi
s
pap
er
,
t
hr
ee
al
gor
i
t
h
m
s
ar
e
us
ed
t
o
t
r
ai
n
t
he
ne
ur
al
net
w
or
k
w
e
i
ght
s
and
t
hr
es
h
ol
ds
,
an
d
t
he
t
r
a
i
n
i
ng
obj
ec
t
i
v
e
f
unc
t
i
o
n
i
s
t
he
m
ean
s
quar
e er
r
or
(
M
S
E
)
.
N
am
el
y
,
(
)
2
1
1
K
i
i
f
y
y
K
=
=
−
∑
(
11
)
W
h
er
e
K
i
s
t
h
e n
um
ber
of
t
r
ai
n
i
ng
s
am
pl
es
,
i
y
i
s
t
he
pr
e
di
c
t
i
v
e
out
put
of
ne
ur
al
net
w
or
k
,
y
is
t
he ac
t
ual
out
put
of
net
w
or
k
.
T
he opt
i
m
i
z
at
i
on
pr
oc
ed
ur
e
bas
ed
on
L
-
M n
eur
a
l
ne
t
w
or
k
i
n t
hi
s
pa
per
i
s
as
f
ol
l
o
w
s
:
S
t
e
p
1
:
I
n
it
ia
li
z
a
t
io
n
T
he c
ont
r
ol
par
am
et
e
r
k
λ
of
L
-
M al
gor
i
t
hm
i
s
i
ni
t
i
al
i
z
ed,
t
h
e m
ax
i
m
u
m
nu
m
ber
o
f
i
t
er
at
i
ons
E
and t
he n
et
w
or
k
al
l
o
w
a
bl
e
er
r
or
ε
ar
e g
i
v
en a
nd t
h
e
w
e
i
ght
s
a
nd t
hr
es
ho
l
ds
of
t
he
neur
a
l
ne
t
w
or
k
v
ec
t
or
W
ar
e
us
ed
as
t
h
e
i
ni
t
i
a
l
s
ol
u
t
i
on.
A
nd
t
he
net
w
or
k
i
s
es
t
abl
i
s
hed
i
n
t
h
i
s
s
t
ep.
S
t
ep
2:
R
and
om
s
el
ec
t
i
on
o
f
t
he t
r
ai
n
i
n
g s
am
pl
es
S
t
ep
3:
C
al
c
u
l
at
i
on
of
t
he
o
bj
ec
t
i
v
e f
unc
t
i
o
n v
al
ue
T
he
m
ean s
quar
e
er
r
or
(
M
S
E
)
f
or
t
r
ai
ni
n
g s
am
pl
es
i
s
c
al
c
ul
a
t
ed b
y
:
(
)
(
)
2
1
1
K
i
i
F
W
y
y
K
=
=
−
∑
(
12)
A
nd
i
t
i
s
s
er
v
e
d as
t
he o
bj
e
c
t
i
v
e
f
unc
t
i
on
.
S
t
ep
4:
J
ud
gm
ent
W
h
et
her
t
he r
es
u
l
t
s
s
at
i
s
f
y
t
he t
er
m
i
nat
i
on c
o
nd
i
t
i
on
i
s
j
udged
b
y
t
he
m
ax
i
m
u
m
n
um
ber
of
i
t
er
at
i
o
ns
E
or
t
he
net
w
or
k
al
l
o
w
ab
l
e
er
r
or
ε
,
i
f
i
t
i
s
,
t
he
pr
oc
edur
e
t
r
ans
f
er
s
t
o
s
t
ep6;
i
f
not
,
t
he pr
oc
e
dur
e
t
r
ans
f
er
s
t
o s
t
ep5
t
o c
on
t
i
n
ue t
he
i
t
er
a
t
i
o
n.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1
6
9
3
-
6
930
T
E
L
KO
M
NI
K
A
V
o
l.
14
,
N
o
.
3,
S
ept
em
ber
2016
:
9
23
–
93
2
928
S
t
ep
5:
I
t
er
at
i
on r
e
ne
w
al
T
he c
al
c
ul
at
i
on
of
t
he
H
es
s
i
an m
at
r
i
x
(
)
(
)
k
HW
of
t
he c
ur
r
ent
s
ol
ut
i
on
and
i
t
s
gr
ad
i
e
nt
(
)
(
)
k
F
W
∇
and t
he r
ene
w
al
of
c
ont
r
ol
par
am
et
er
k
λ
ar
e done i
n
t
hi
s
s
t
ep.
T
he w
ei
g
ht
s
an
d
t
hr
es
hol
ds
ar
e
upd
at
e
d ac
c
or
di
n
g t
o
t
he
i
t
er
at
i
v
e f
or
m
u
l
a (
6)
a
nd
pr
oc
edur
e t
r
ans
f
er
s
t
o s
t
ep3
.
S
t
ep
6:
O
ut
put
A
s
et
of
w
ei
ght
s
and t
hr
es
hol
ds
ar
e out
p
ut
t
e
d as
t
he opt
i
m
al
s
ol
ut
i
o
n,
an
d t
he
pr
oc
edur
e ends
.
T
he f
l
ow
c
har
t
i
s
s
ho
w
n i
n
F
i
gur
e 2.
F
i
gur
e
2
.
T
he f
l
o
w
c
h
ar
t
of
B
P
neur
a
l
n
et
w
or
k
t
r
ai
ni
ng
pr
oc
es
s
b
y
L
-
M a
l
gor
i
t
hm
4
.
E
n
g
i
n
e
T
e
s
t S
y
s
t
e
m
a
n
d
E
l
e
c
tr
i
c
C
o
n
tr
o
l
U
n
i
t
4.
1.
E
n
g
i
n
e T
est
S
yst
em
T
he h
y
dr
og
en
-
f
uel
e
d eng
i
n
e benc
h t
es
t
i
s
m
odi
f
i
ed b
y
a
s
i
ngl
e c
y
l
i
n
der
f
our
s
t
r
o
k
e v
al
v
e
w
at
er
c
ool
ed e
ng
i
ne,
and
i
t
s
par
am
et
er
s
ar
e as
s
h
o
w
n i
n T
abl
e 1 b
el
o
w
.
T
he t
es
t
s
y
s
t
em
c
ons
i
s
t
s
of
t
he h
y
dr
og
en s
upp
l
y
s
y
s
t
em
,
ai
r
s
uppl
y
s
y
s
t
em
,
ex
haus
t
s
y
s
t
em
,
i
g
ni
t
i
on s
y
s
t
em
,
el
ec
t
r
on
i
c
c
ont
r
o
l
un
i
t
,
v
ar
i
ous
t
es
t
s
ens
o
r
s
and
s
i
gn
a
l
c
ond
i
t
i
on
i
ng
c
i
r
c
u
i
t
,
c
om
put
er
m
oni
t
or
i
n
g
s
y
s
t
em
,
d
y
n
am
o
m
et
er
,
heat
-
di
s
s
i
p
at
i
on m
odul
e
and
v
ar
i
ous
c
ont
r
o
l
v
al
v
e s
y
s
t
em
s
.
T
he t
es
t
s
ens
or
s
i
nc
l
ude:
t
h
e h
y
dr
ogen
f
l
o
w
s
e
ns
or
,
a
i
r
f
l
o
w
s
ens
or
,
i
nt
ak
e pr
es
s
ur
e
s
ens
or
,
s
pee
d
s
ens
or
,
t
hr
ot
t
l
e
p
os
i
t
i
on
s
e
ns
or
,
c
ool
i
ng
w
at
er
t
em
per
at
ur
e
s
ens
or
,
et
c
.
,
w
i
t
h
t
he
abi
l
i
t
y
t
o
c
ar
r
y
on
a
c
om
pr
ehens
i
v
e
t
es
t
i
ng
a
nd
m
oni
t
or
i
ng
of
t
he
h
y
dr
oge
n
-
f
uel
ed
en
gi
ne
under
m
ul
t
i
pl
e
w
or
k
i
ng c
ondi
t
i
ons
.
B
y
us
i
ng t
he t
es
t
benc
h t
o c
a
l
i
br
at
e t
he i
gni
t
i
o
n dat
a,
c
om
bi
n
ed
w
i
t
h
t
he
i
nt
e
l
l
i
g
ent
al
g
or
i
t
hm
pr
opos
ed i
n
t
h
i
s
pap
er
,
t
h
e
w
h
ol
e c
o
nd
i
t
i
on c
a
l
i
br
at
i
on a
nd o
pt
i
m
i
z
at
i
o
n
c
ont
r
ol
of
t
he e
ng
i
ne
i
g
ni
t
i
o
n s
y
s
t
em
c
an be c
ar
r
i
e
d o
u
t
.
T
abl
e 1.
T
he p
ar
am
et
er
s
of
t
he
h
y
dr
oge
n
-
f
uel
e
d en
gi
n
e
C
y
l
i
nder
di
am
e
t
er
（
m
m
）
P
i
s
t
on
t
r
av
el
（
m
m
）
D
i
s
pl
ac
e
m
ent
（
L）
C
om
pr
e
s
s
i
on
ra
t
i
o
M
a
x
i
m
u
m
pow
er
（
KW
）
M
a
xi
m
u
m
s
peed
94
85
0.
5899
9.
7
30
6000
4
.
2
.
T
h
e
E
l
e
c
tr
i
c
C
o
n
tr
o
l
U
n
i
t
I
n
t
he
pr
oc
es
s
of
engi
ne
r
u
nni
ng,
t
h
e
el
ec
t
r
oni
c
c
ont
r
o
l
un
i
t
(
E
C
U
)
r
ea
ds
t
he
op
e
r
at
i
on
par
am
et
er
s
f
r
o
m
t
he
v
ar
i
o
us
s
ens
or
s
[
11]
,
af
t
er
t
he
ana
l
y
s
i
s
a
nd
pr
oc
es
s
i
ng i
s
c
ar
r
i
ed
out
,
an
i
ns
t
r
uc
t
i
o
n i
s
i
s
s
ued t
o e
ac
h en
gi
n
e ac
t
u
at
or
t
o c
ont
r
o
l
t
he o
per
at
i
o
n of
t
he
eng
i
n
e
.
A
t
t
hi
s
s
t
ag
e,
t
he
c
om
pl
ex
i
t
y
of
t
he
eng
i
n
e
el
ec
t
r
on
i
c
c
ont
r
ol
u
ni
t
de
t
er
m
i
nes
t
he
degr
ee
of
adapt
at
i
o
n
of
t
he
B
egi
n
I
ni
t
i
al
i
z
e t
he
c
ont
r
ol
par
a
m
et
er
k
λ
I
ni
t
i
al
i
z
e t
he w
ei
ght
s
and
t
hr
es
hol
d
s
W
，
bui
l
d t
he
net
w
or
k
S
el
ec
t
t
he t
r
ai
ni
ng
s
a
m
pl
e
s
r
andom
l
y
C
al
c
ul
at
e
t
he obj
e
c
t
i
v
e f
un
c
t
i
on
（
MS
E
）
U
pdat
e t
he
c
ont
r
ol
par
a
m
et
er
k
λ
U
pdat
e t
he op
t
i
m
al
w
ei
ght
s
and
t
h
r
es
hol
ds
E
nd j
udg
m
en
t
E
nd
NO
O
ut
put
a
s
et
o
f
w
ei
ght
s
and
t
hr
es
h
ol
ds
a
s
Y
ES
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
K
A
I
S
S
N
:
1
693
-
6
930
O
pt
i
m
i
z
at
i
on
of
H
y
dr
o
gen
-
f
uel
ed E
ng
i
n
e
I
gn
i
t
i
o
n T
i
mi
ng
B
a
s
ed
o
n L
-
M
N
eur
a
l
…
(
Li
j
un W
ang
)
929
eng
i
ne i
n v
ar
i
ous
c
on
di
t
i
on
s
,
and i
t
s
r
ol
e i
s
e
qui
v
a
l
e
n
t
t
o t
he br
a
i
n of
hum
an
[
12
]
.
T
her
ef
or
e,
af
t
er
t
he c
om
pl
et
i
o
n of
t
he
h
y
dr
o
gen
-
f
ue
l
ed
eng
i
ne
o
pt
i
m
i
z
at
i
on c
o
nt
r
ol
al
gor
i
t
h
m
,
t
he c
ont
r
ol
al
g
or
i
t
hm
needs
t
o be w
r
i
t
t
en i
nt
o t
he el
ec
t
r
on
i
c
c
ont
r
ol
un
i
t
.
A
c
c
or
di
ng t
o t
h
e i
g
ni
t
i
on s
y
s
t
em
c
ont
r
ol
s
t
r
at
eg
y
pr
op
os
e
d
i
n
t
hi
s
pa
per
,
t
h
e
h
y
dr
oge
n
-
f
uel
ed
eng
i
n
e
i
g
n
i
t
i
on
M
A
P
i
s
m
anuf
ac
t
ur
ed and i
m
por
t
e
d i
nt
o E
C
U
.
T
he s
c
hem
at
i
c
di
a
gr
am
of
t
he el
ec
t
r
on
i
c
c
ont
r
ol
uni
t
i
s
s
ho
w
n
i
n F
i
gur
e 3.
F
i
gur
e
3
.
T
he e
l
ec
t
r
on
i
c
c
o
nt
r
ol
uni
t
5
.
E
x
p
e
r
i
m
e
n
t S
i
m
u
l
a
ti
o
n
5
.1
. S
im
u
la
t
io
n
T
he B
P
n
eur
a
l
ne
t
w
or
k
opt
i
m
i
z
at
i
on m
odel
bas
e
d on
L
-
M al
g
or
i
t
hm
i
s
es
t
abl
i
s
h
ed
i
n t
h
i
s
paper
,
an
d
c
om
par
ed
w
i
t
h
t
he
P
o
w
el
l
ne
ur
al
ne
t
w
or
k
al
g
or
i
t
hm
and
t
he
t
r
adi
t
i
o
na
l
B
P
a
l
g
or
i
t
hm
b
y
us
i
n
g
MA
T
LA
B
n
eur
a
l
net
w
or
k
t
oo
l
box
f
or
s
i
m
ul
at
i
on
c
al
c
u
l
at
i
ng
.
T
he
p
ar
a
m
et
er
s
of
ne
ur
al
net
w
or
k
ar
e s
et
t
o:
t
he
num
ber
of
i
n
put
l
a
y
er
ne
ur
ons
2
n
=
,
t
he
num
ber
of
hi
d
den
l
a
y
er
neur
o
n
s
5
q
=
,
t
he
n
um
ber
of
out
put
l
a
y
e
r
neur
o
n
1
m
=
,
t
he
m
ax
i
m
u
m
num
ber
of
i
t
er
at
i
o
ns
100
E
=
and
t
he net
w
or
k
al
l
o
w
a
bl
e er
r
or
3
10
ε
−
=
.
T
he par
am
et
er
s
o
f
L
-
M al
g
or
i
t
hm
ar
e s
et
t
o:
t
h
e i
ni
t
i
a
l
c
ont
r
ol
par
am
et
er
10
k
λ
=
and
t
he
num
ber
of
s
a
m
pl
es
84
K
=
.
5.
2
.
R
esu
l
t
A
n
a
l
y
si
s
I
n
or
der
t
o
v
er
i
f
y
t
h
e
opt
i
m
i
z
at
i
on
ef
f
ec
t
i
v
enes
s
of
t
he
L
-
M
neur
a
l
ne
t
w
or
k
al
gor
i
t
hm
f
o
r
i
gn
i
t
i
on
t
i
m
i
ng
of
t
he
h
y
dr
o
gen
-
f
uel
ed
en
gi
ne,
f
i
r
s
t
l
y
,
t
he
t
r
ai
ni
ng
s
am
pl
es
ar
e
us
ed
t
o
t
r
a
i
n
t
h
e
net
w
or
k
,
and
t
h
en
t
h
e
t
es
t
s
a
m
pl
es
ar
e
us
ed
t
o
t
es
t
t
he
t
r
a
i
n
ed
net
w
or
k
.
T
he
r
es
ul
t
s
s
ho
w
t
h
at
t
he
L
-
M n
eur
a
l
n
et
w
or
k
al
g
or
i
t
hm
has
hi
gh
ac
c
ur
ac
y
a
nd h
i
g
h s
pe
ed
w
he
n us
e
d t
o op
t
i
m
i
z
e
t
he
i
gn
i
t
i
on
t
i
m
i
ng,
w
h
i
c
h
c
o
m
pl
et
el
y
m
eet
s
t
h
e
des
i
g
n
r
equ
i
r
em
ent
s
.
I
t
s
opt
i
m
i
z
at
i
on
ef
f
ec
t
i
s
s
uper
i
or
t
o
t
he
t
r
a
di
t
i
o
nal
B
P
n
et
w
or
k
al
gor
i
t
hm
,
and
c
om
par
ed
w
i
t
h
t
he
P
o
w
e
l
l
al
gor
i
t
hm
w
hi
c
h
i
s
m
or
e
s
ui
t
abl
e
f
or
c
om
pl
ex
f
unc
t
i
ons
,
i
t
a
l
s
o
has
a
hi
g
her
c
on
v
er
g
enc
e
s
p
eed
and
ac
c
ur
ac
y
.
W
h
en t
he i
t
er
at
i
on n
um
ber
s
at
i
s
f
i
es
t
he t
er
m
i
nat
i
on c
o
ndi
t
i
o
n,
t
h
e m
ean s
quar
e e
r
r
or
of
t
he L
-
M
al
g
or
i
t
hm
i
s
0.
0
0
28,
t
he
P
o
w
el
l
a
l
g
or
i
t
hm
i
s
0.
0
057
8
5 an
d t
h
e B
P
a
l
g
or
i
t
hm
i
s
0.
16
56.
T
he
c
onv
er
g
enc
e c
ur
v
es
of
t
he
t
hr
ee a
l
g
or
i
t
hm
s
ar
e s
ho
w
n
i
n
F
i
gur
e
4,
5 an
d 6.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1
6
9
3
-
6
930
T
E
L
KO
M
NI
K
A
V
o
l.
14
,
N
o
.
3,
S
ept
em
ber
2016
:
9
23
–
93
2
930
F
i
gur
e
4.
T
he
c
on
v
er
g
enc
e
c
ur
v
e of
L
-
M
al
g
or
i
t
hm
F
i
gur
e
5.
T
he
c
on
v
er
g
enc
e
c
ur
v
e of
P
o
w
e
l
l
a
lg
o
r
it
h
m
F
i
gur
e
6.
T
he
c
on
v
er
g
enc
e
c
ur
v
e of
B
P
a
l
gor
i
t
hm
T
he
abs
ol
ut
e
er
r
or
c
ur
v
es
bet
w
e
en
t
he
s
i
m
ul
at
e
d
and
ac
t
ual
v
a
l
u
es
of
al
l
s
am
pl
es
ar
e
s
ho
w
n
i
n
F
ig
u
r
e
7,
8
a
nd
9
.
A
m
ong
t
hem
,
t
he
m
ax
i
m
um
abs
ol
ut
e
er
r
or
of
t
he
L
-
M
al
g
or
i
t
hm
i
s
0.
24
564,
t
h
e P
o
w
e
l
l
a
l
gor
i
t
hm
i
s
0.
2679
,
a
nd t
he
B
P
al
gor
i
t
hm
i
s
1.
447
.
T
hi
s
s
ho
w
s
t
h
at
t
he
opt
i
m
i
z
at
i
o
n ef
f
ec
t
of
L
-
M al
gor
i
t
hm
i
s
t
he
bes
t
w
i
t
h
t
he
l
eas
t
er
r
or
and
hi
ghes
t
s
p
e
ed.
F
i
gur
e
7.
T
he abs
ol
u
t
e er
r
o
r
c
ur
v
e of
L
-
M
al
g
or
i
t
hm
F
i
gur
e
8.
T
he abs
ol
u
t
e er
r
o
r
c
ur
v
e
of
P
o
w
el
l
al
g
or
i
t
hm
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
K
A
I
S
S
N
:
1
693
-
6
930
O
pt
i
m
i
z
at
i
on
of
H
y
dr
o
gen
-
f
uel
ed E
ng
i
n
e
I
gn
i
t
i
o
n T
i
mi
ng
B
a
s
ed
o
n L
-
M
N
eur
a
l
…
(
Li
j
un W
ang
)
931
F
i
gur
e
9.
T
he abs
ol
u
t
e er
r
o
r
c
ur
v
e of
B
P
a
l
gor
i
t
hm
T
he
c
al
c
ul
at
ed
r
es
u
l
t
s
of
MA
P
di
agr
am
of
h
y
dr
og
en
-
f
uel
e
d
en
gi
n
e
i
gn
i
t
i
on
s
y
s
t
e
m
bas
ed
on L
-
M ne
ur
al
net
w
o
r
k
opt
i
m
i
z
at
i
on
al
gor
i
t
hm
ar
e s
ho
w
n
i
n
F
i
g
ur
e 1
0.
F
i
gur
e
10.
T
he c
a
l
c
ul
at
ed
i
gni
t
i
o
n M
A
P
di
agr
am
bas
ed on
L
-
M n
eur
a
l
n
et
w
or
k
al
gor
i
t
hm
6
.
C
o
n
c
l
u
s
i
o
n
T
he
i
gn
i
t
i
on
t
i
m
i
ng
op
t
i
m
i
z
a
t
i
on
m
odel
of
t
he
h
y
dr
og
en
-
f
uel
ed
e
ng
i
ne
bas
e
d
o
n
t
h
e
L
-
M
neur
a
l
net
w
or
k
al
gor
i
t
hm
i
s
c
ons
t
r
uc
t
ed i
n t
hi
s
pa
per
.
I
t
has
hi
gh pr
ec
i
s
i
o
n,
h
i
g
h
c
onv
er
genc
e
s
peed,
a
s
i
m
pl
e
m
odel
an
d ot
her
ad
v
an
t
ag
es
i
n t
he
net
w
or
k
t
r
ai
ni
n
g
pr
oc
es
s
.
I
t
c
o
m
bi
nes
t
he
adv
ant
ages
of
N
e
w
t
on m
et
ho
d an
d gr
ad
i
e
nt
m
et
hod,
an
d a
v
oi
ds
t
he c
h
ar
ac
t
er
i
s
t
i
c
of
l
oc
al
ex
t
r
em
u
m
.
C
o
m
par
ed w
i
t
h
t
he
t
r
ad
i
t
i
ona
l
B
P
al
gor
i
t
h
m
and P
o
w
el
l
a
l
gor
i
t
hm
,
i
t
has
a
bet
t
er
opt
i
m
i
z
at
i
o
n ef
f
ec
t
.
T
he c
al
i
br
at
i
on
and
opt
i
m
i
z
a
t
i
o
n c
ont
r
ol
of
t
he i
gni
t
i
o
n
t
i
m
i
ng of
t
he
hy
d
r
og
en
-
f
uel
ed
eng
i
n
e un
der
w
ho
l
e
w
or
k
i
ng c
ond
i
t
i
o
ns
i
s
r
eal
i
z
ed.
T
he op
t
i
m
i
z
a
t
i
on m
odel
has
a
v
er
y
s
t
r
on
g
pr
ac
t
i
c
ab
i
l
i
t
y
,
and
i
t
i
s
s
i
g
ni
f
i
c
ant
i
n
t
he
e
x
per
i
m
ent
al
s
t
ud
y
of
t
he
h
y
dr
oge
n
-
f
uel
ed
eng
i
ne.
A
c
k
n
o
w
l
e
d
g
e
m
e
n
ts
T
hi
s
w
or
k
w
as
s
uppor
t
ed
b
y
P
l
a
n f
or
S
c
i
ent
i
f
i
c
I
nn
o
v
at
i
on T
al
e
nt
of
H
enan P
r
ov
i
nc
e
(
1641
005
100
18)
,
H
A
S
T
I
T
(
14H
A
S
T
I
T
001)
and Z
hen
g
z
h
ou
Me
as
ur
i
n
g &
C
o
nt
r
ol
T
ec
hno
l
og
y
and I
ns
t
r
um
ent
at
i
o
ns
K
e
y
L
abor
at
or
y
(
1
21
P
Y
F
Z
X
18
1)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN
:
1
6
9
3
-
6
930
T
E
L
KO
M
NI
K
A
V
o
l.
14
,
N
o
.
3,
S
ept
em
ber
2016
:
9
23
–
93
2
932
R
ef
er
en
ces
[1
]
S
V
er
hel
s
t
.
R
e
c
en
t
P
r
ogr
es
s
i
n t
he
U
s
e
of
H
y
dr
oge
n A
s
a
F
uel
f
or
I
n
t
er
na
l
C
o
m
bu
s
t
i
o
n
E
egi
n
es
.
I
nt
e
r
nat
i
onal
J
ou
r
nal
of
H
y
d
r
oge
n E
n
er
gy
.
2014;
39(
2)
:
10
71
-
108
5
.
[2
]
Z
hang
B
o,
J
i
C
han
gw
ei
,
W
a
n
g
S
huof
eng,
X
i
ao
Y
uc
he
n.
I
nv
es
t
i
g
at
i
o
n
o
n
t
h
e
C
o
l
d
S
t
ar
t
C
h
ar
ac
t
er
i
s
t
i
c
s
of
a H
y
dr
og
en
-
enr
i
c
hed M
et
hanol
E
ng
i
ne.
I
nt
er
nat
i
o
nal
J
o
ur
nal
of
H
y
d
r
oge
n
E
ner
gy
.
2014;
39(
36)
:
14466
-
144
71.
[3
]
JM
G
om
es
A
nt
u
nes
,
R
M
i
k
al
s
en,
A
P
R
os
k
i
l
l
y
.
A
n E
x
per
i
m
ent
a
l
S
t
udy
of
a D
i
r
e
c
t
I
n
j
ec
t
ion
C
om
pr
e
s
s
i
on
I
g
ni
t
i
on
H
y
dr
og
en
E
ngi
ne.
I
nt
er
na
t
i
ona
l
J
our
n
a
l
of
H
y
dr
ogen
E
ner
gy
.
2009;
3
4(
15)
:
65
16
-
6522.
[4
]
Li
u
X
i
nghu
a,
Li
u
F
us
hui
,
Z
ho
u
Lei
,
S
u
n,
B
ai
gan
g,
S
c
h
oc
k
H
ar
ol
d.
B
a
c
k
f
i
r
e
P
r
e
di
c
t
i
o
n
i
n
a
M
ani
f
ol
d
I
nj
e
c
t
i
o
n H
y
dr
ogen I
nt
er
na
l
C
o
m
bu
s
t
i
o
n E
ng
i
ne.
I
nt
er
na
t
i
ona
l
J
our
nal
o
f
H
y
d
r
og
en E
n
er
gy
.
20
08;
33(
1
4)
:
3847
-
3
855.
[5
]
Y
adav
V
i
nod
S
i
ngh
,
S
har
m
a
D
i
l
i
p
,
S
o
ni
S
L.
P
er
f
or
m
anc
e
and
C
om
bus
t
i
on
A
na
l
y
s
i
s
of
H
y
dr
ogen
-
f
uel
l
ed C
.
I
.
E
ngi
ne w
i
t
h E
G
R
.
I
nt
er
n
at
i
o
nal
J
our
n
al
o
f
H
y
dr
ogen
E
ne
r
gy
.
2015;
40(
2)
:
4
382
-
439
1
.
[6
]
S
ubr
ot
o,
I
m
a
m
M
uc
h
I
bnu,
S
e
l
am
a
t
,
A
l
i
.
P
l
agi
ar
i
s
m
D
et
e
c
t
i
o
n
T
hr
ough
I
nt
er
net
U
s
i
ng
H
y
b
r
i
d
A
r
t
i
f
i
c
i
al
N
eur
al
N
et
w
or
k
and S
upp
or
t
V
ec
t
or
s
M
ac
hi
ne.
T
el
k
om
ni
k
a
(
T
e
l
ec
om
m
uni
c
a
t
i
on
C
om
put
i
n
g
E
l
ec
t
r
oni
c
s
and
C
o
nt
r
o
l
)
.
20
14
;
12(
1)
:
209
-
218.
[7
]
B
al
j
i
t
S
i
ng
h
K
hehr
aa,
A
m
ar
P
ar
t
ap
S
i
ngh
P
h
ar
w
ahab.
C
l
a
s
s
i
f
i
c
at
i
on
o
f
C
l
us
t
er
e
d
M
i
c
r
oc
al
c
i
f
i
c
at
i
ons
us
i
n
g M
LF
F
B
P
-
AN
N
a
n
d
SVM
.
E
gy
pt
i
a
n I
n
f
or
m
at
i
c
s
J
our
nal
.
2016;
1
7(
1)
:
11
-
20
.
[8
]
O
s
am
a
h
B
as
heer
S
huk
ur
,
M
uham
m
ad
H
i
s
y
am
Lee.
D
ai
l
y
W
i
n
d
S
p
eed
F
or
e
c
as
t
i
n
g
T
hr
ough
H
y
br
i
d
KF
-
A
N
N
M
odel
B
a
s
ed on
A
R
I
M
A
.
R
ene
w
a
bl
e
E
ne
r
gy
.
2015;
76:
63
7
-
64
7.
[9
]
A
dam
P
P
i
ot
r
ow
s
k
i
,
J
ar
os
ł
aw
J
N
api
or
k
ow
s
k
i
.
O
pt
i
m
i
z
i
n
g N
eur
al
N
e
t
w
or
k
s
fo
r
R
i
v
er
F
l
ow
F
or
ec
as
t
i
n
g
–
E
v
ol
ut
i
o
nar
y
C
om
put
at
i
on
M
et
hods
v
e
r
s
us
t
he Lev
e
nber
g
–
M
ar
quar
d
t
A
ppr
oac
h.
J
our
nal
of
H
y
dr
ol
ogy
.
2011;
40
7
(1
-
4)
:
12
-
27.
[1
0
]
CA
C
hr
i
s
t
odou
l
ou
a,
V
Vi
t
a
b
,
L
E
k
ono
m
ou
b,
G
E
C
h
at
z
ar
ak
i
s
b
,
I
A
S
t
at
hop
ul
o
s
a.
A
ppl
i
c
at
i
o
n
of
P
ow
el
l
’
s
O
pt
i
m
i
z
at
i
on M
et
hod t
o S
ur
g
e
A
r
r
es
t
er
C
i
r
c
u
i
t
M
odel
s
’
P
ar
a
m
et
er
s
.
E
ner
gy
.
2010;
35(
8)
:
3
37
5
-
3380
.
[1
1
]
T
r
i
w
i
y
at
no
Ari
s
,
N
u
h
M
oham
m
ad,
S
ant
o
s
o
A
ri
,
N
y
om
a
n
S
ut
ant
r
a
.
E
ng
i
ne
T
or
qu
e
C
ont
r
ol
of
S
par
k
I
gni
t
i
on E
n
gi
n
e U
s
i
n
g F
uz
z
y
G
ai
n S
c
hedu
l
i
n
g.
T
el
k
o
m
n
ik
a
(
T
el
ec
om
m
uni
c
a
t
i
on
C
om
put
i
n
g E
l
ec
t
r
oni
c
s
and C
on
t
r
ol
)
.
201
2;
1
0(
1)
:
83
-
90.
[1
2
]
T
K
r
us
e,
S
K
ur
z
,
T
La
ng.
M
oder
n S
t
at
i
s
t
i
c
al
M
odel
i
n
g a
nd E
v
ol
ut
i
on
ar
y
O
pt
i
m
i
z
at
i
o
n M
et
hods
f
or
t
h
e
B
r
oad U
s
e i
n
E
C
U
C
al
i
br
at
i
on.
M
uni
c
h
.
20
10;
1
:
73
9
-
743
.
Evaluation Warning : The document was created with Spire.PDF for Python.