Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
23
,
No.
1
,
Ju
ly
2021
, p
p.
33
8
~
3
44
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v
23
.i
1
.
pp
338
-
344
338
Journ
al h
om
e
page
:
http:
//
ij
eecs.i
aesc
or
e.c
om
A
ne
w p
aram
eter in thre
e
-
t
er
m
conjug
ate gradi
ent a
l
gorith
ms
for unc
on
stra
in
ed
optimization
Alaa S
aad
Ahm
ed
1
,
His
ha
m
M
.
Kh
udh
ur
2
, Mohamme
d
S. Najmuldee
n
3
1
Com
pute
r
Scie
n
ce
Dep
art
m
ent,
Coll
ege of
Educat
ion
for
pur
e
sc
i
enc
es,
U
nive
rsi
t
y
of
Mos
ul, Mosul,
Ir
aq
2
Mathe
m
at
i
cs
Depa
rtment
,
Co
ll
e
ge
of
Com
put
er S
ci
ence and
Mat
hemati
cs,
Unive
rsit
y
of
Mos
ul,
Mos
ul,
Ira
q
3
Ministr
y
of Edu
ca
t
ion
,
Kirkuk, I
raq
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
N
ov
1,
2020
Re
vised A
pr
1
9
, 2
021
Accepte
d
J
un
1
8
, 202
1
In
thi
s
stud
y
,
we
develop
a
d
iffe
ren
t
par
amet
er
of
thr
ee
te
r
m
conj
ugate
gra
die
n
t
kind,
th
is
sche
m
e
depe
n
ds
princ
ipa
l
l
y
o
n
pure
conj
uga
c
y
cond
it
ion
(PCC
),
W
her
ea
s,
the
con
juga
c
y
condi
ti
on
(P
CC)
is
an
important
condi
ti
on
in
unconstra
in
ed
non
-
li
ne
ar
opti
m
i
za
t
ion
in
gene
r
a
l
and
in
conj
ug
at
e
gra
d
ie
n
t
m
et
hods
in
par
ticula
r
.
The
propo
sed
m
et
hod
bec
o
m
es
conve
rge
d,
and
sati
sf
y
condi
ti
ons
d
esc
ent
prope
r
t
y
b
y
assum
ing
som
e
h
y
poth
esis,
Th
e
num
eri
c
al
res
ult
s
display
the
eff
ective
n
ess
of
the
ne
w
m
et
hod
for
solving
te
s
t
unconstra
in
ed
non
-
li
ne
ar
opti
m
i
za
t
ion
proble
m
s
compare
d
to
oth
er
conj
uga
te
gra
die
n
t
a
lgori
th
m
s such
as
fle
tcher
and revees
(
FR
)
al
gorit
hm
a
nd
three
t
erm
fle
t
che
r
and
rev
e
es
(TT
FR
)
al
gor
it
hm
.
The
num
er
ic
a
l
result
s
dem
onstrat
e
th
e
eff
icac
y
of
the
suggested
m
et
hod
for
solving
t
est
unconstra
in
e
d
nonli
nea
r
opti
m
iz
ation
proble
m
s
from
whe
re
a
num
ber
of
i
te
ra
ti
ons
and
eva
luation
o
f
func
ti
on
and
A com
par
ison o
f
th
e
ti
m
e ta
k
en
to
pe
rform
the
fun
ct
i
ons.
Ke
yw
or
ds:
Algorithm
Conj
ug
at
e
grad
ie
nt
pure c
onju
gacy
Un
c
onstrai
ne
d op
ti
m
iz
ation
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Hish
am
M. Kh
udhur
Ma
them
a
ti
cs D
epar
tm
ent
Coll
ege
of
C
om
pu
te
r
Scie
nc
e
an
d
Ma
them
at
ic
s
U
niv
e
rsity
o
f
Mosu
l,
Iraq
Em
a
il
:
hish
am
892020
@uom
os
ul
.edu.i
q
1.
INTROD
U
CTION
Re
searche
rs
ha
ve
stu
died
the
pro
blem
of
unr
est
rict
ed
i
m
pr
ovem
ent
as
a
m
at
te
r
of
fi
nd
i
ng
a
so
luti
on
to the m
ini
m
izati
on
of the
rea
l functi
on
(
)
.
∈
(
)
(1)
Wh
e
reas
(
)
a
deri
vative f
unct
io
n
at
least
on
ce
.
Conj
ug
at
e
gra
dient
(CG
)
al
gorithm
s
are
i
m
po
rtant
to
so
lve
for
(1)
prob
le
m
us
ing
the
fo
ll
owin
g
it
erati
ve
m
et
ho
d:
+
1
=
+
=
0
,
1
,
2
(2)
Wh
e
reas
cal
cu
la
te
s the step
size i
n
ei
the
r
e
xa
ct
ly
o
r
ine
xactl
y l
ine searc
h u
sing t
he follo
w
ing
relat
ion
:
(
+
)
=
≥
0
(
+
)
(3)
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A n
ew
par
am
et
er in t
hr
ee
-
te
r
m
c
on
j
ugate
grad
ie
nt a
l
gorit
hm
s fo
r
un
c
onst
ra
ine
d…
(
Ala
a S
aad Ah
me
d
)
339
+
1
is a sea
rch dir
ect
ion
a
nd it
is
known
as t
he
f
ollow
i
ng for
m
ula:
1
=
−
1
=
1
+
1
=
−
+
1
+
≥
1
(4)
+
1
is a
vector m
atr
ix
of fu
nction
, and
is a C
G m
et
ho
d
pa
ram
et
er.
Be
low
a
re
par
a
m
et
ers
for
s
ome
conju
gate
gr
a
dient alg
ori
thm
s:
is cal
culat
ed wit
h
the
searc
h direct
io
n
+
1
in t
he
foll
ow
i
ng form
ulas:
1
=
−
1
,
=
+
1
−
1
-
+
1
=
−
+
1
+
+
1
+
1
[1]
2
-
+
1
=
−
+
1
+
+
1
[2]
See
[
3]
-
[
8]
.
Ther
e
a
re
thre
e
-
te
rm
con
juga
te
gr
adie
nt
m
et
hods
f
or
th
re
e
par
am
et
ers
(F
R,
PR,
HS)
pro
po
se
d
by
Zha
ng
[
9]
.
T
he
se
thr
ee
m
et
ho
ds
al
ways
ac
hieve
re
gr
es
sio
n
pro
per
ty
.
Be
low
is
the
sear
ch
directi
on
for
s
om
e
three
-
te
rm
co
nju
gate
d gr
a
dien
t
m
et
ho
ds
:
1)
The
c
onju
gate
gr
a
dient m
e
thod
(F
R)
with
three
-
te
rm
is k
no
wn as:
+
1
=
−
+
1
+
−
(
1
)
+
1
Wh
e
reas
(
1
)
=
+
1
2)
The
c
onju
gate
gr
a
dient m
et
hod (PR) wit
h t
hree
-
te
rm
is k
no
wn as:
+
1
=
−
+
1
+
−
(
2
)
Wh
e
reas
(
2
)
=
+
1
3)
The
c
onju
gate
gr
a
dient m
et
hod (H
S
) wit
h
t
hree
-
te
rm
is k
no
wn as:
+
1
=
−
+
1
+
−
(
3
)
Wh
e
reas
(
3
)
=
+
1
+
1
We
no
ti
ce t
hat
these m
et
ho
ds
al
ways achie
ve
the
fo
ll
owin
g pro
per
ty
:
=
‖
‖
2
<
0
∀
Her
e
the
regres
sion p
rope
rty
is achie
ved w
it
h
=
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
23
, N
o.
1
,
Ju
ly
2021
:
3
3
8
-
3
4
4
340
Of
te
n
the
r
esea
rch
e
r need
s eit
her exactl
y o
r
i
nex
act
ly
li
ne
s
earch
whe
n
stu
dying co
nver
ge
nce a
nd
app
ly
in
g
th
e C
G
m
et
ho
d. Lik
e the str
ong
Wo
lf c
onditi
ons.
The
st
ron
g Wol
f
co
ndit
ion
s
ar
e to
fin
d
k
(
+
)
≤
(
)
+
(5)
|
(
+
)
|
≤
−
(6)
0
<
<
<
1
are c
onsta
nts a
ccordin
g
t
o
Li
and
W
ei
j
un
[
3]
, [9],
[10]
.
In
S
e
c
t
i
o
n
2
,
w
e
p
r
e
s
e
n
t
t
h
e
d
e
r
i
v
a
t
i
o
n
o
f
t
h
e
n
e
w
m
e
t
h
o
d
u
s
i
n
g
t
h
e
F
R
c
o
n
j
u
g
a
t
e
g
r
a
d
i
e
n
t
m
e
t
h
o
d
w
i
t
h
t
h
r
e
e
-
t
e
r
m
.
I
n
S
e
c
t
i
o
n
3
,
w
e
e
x
p
l
a
i
n
t
h
e
r
e
g
r
e
s
s
i
o
n
o
f
t
h
e
n
e
w
m
e
t
h
o
d
.
I
n
S
e
c
t
i
o
n
4
,
w
e
e
x
p
l
a
i
n
t
h
e
a
b
s
o
l
u
t
e
c
o
n
v
e
r
g
e
n
c
e
o
f
t
h
e
n
e
w
i
m
p
r
o
v
e
d
a
l
g
o
r
i
t
h
m
.
I
n
S
e
c
t
i
o
n
5
,
t
h
e
n
u
m
e
r
i
c
a
l
r
e
s
u
l
t
s
o
f
t
h
e
p
r
o
p
o
s
e
d
a
l
g
o
r
i
t
h
m
a
r
e
p
r
e
s
e
n
t
e
d
,
a
n
d
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
t
h
e
n
e
w
i
m
p
r
o
v
e
d
a
l
g
o
r
i
t
h
m
i
s
c
o
m
p
a
r
e
d
w
i
t
h
o
t
h
e
r
a
l
g
o
r
i
t
h
m
s
i
n
t
h
e
s
a
m
e
f
i
e
l
d
.
2.
IMP
ROVIN
G
THE MET
H
OD OF
CON
JUG
ATED
GRADIE
NT F
R WIT
H TH
RE
E
-
TE
RM
+
1
=
−
+
1
+
−
+
1
Wh
e
re
∈
[
0
,
1
]
an
d usi
ng p
ure c
onjuga
cy
co
ndit
io
n
[
11
]
+
1
=
−
+
1
+
−
+
1
=
0
+
1
=
−
+
1
+
=
−
+
+
1
(
7)
+
1
=
−
+
1
+
−
+
1
(
8)
Algori
th
m
:
The
c
onju
gate
gr
a
dient m
et
hod (FR) a
lg
or
it
hm
w
it
h
i
m
pr
oved
th
ree
-
te
rm
:
Step
1: Let
0
is
an
init
ia
l val
ue
,
put
0
=
−
0
,
>
0
,
=
0
.
Step
2: D
et
e
rm
ine the le
ngth
of the ste
p
>
0
ach
ie
ves
the
Wo
lf
e co
nd
it
io
ns
(5),
(6).
Step
3:
Dete
rm
ine
+
1
=
+
. I
f
‖
+
1
‖
<
then st
oppe
d.
Step
4:
Dete
rm
ine
+
1
,
from
(
7)
and g
e
ne
rate di
recti
on fro
m
(
8).
Step
5: P
ut
=
+
1
. Go
to
step
2.
3.
REGR
E
SSIO
N
P
ROPE
RT
Y
O
F THE
N
EW FO
RMU
LA
We
will
m
entio
n
the
pro
of
of
the
su
f
fici
ent
desce
nt
pro
pert
y
fo
r
the
co
nj
ug
at
e
gra
dient
m
et
ho
d
(
FR)
al
gorithm
fo
rm
ula
with
i
m
pr
oved
th
ree
-
te
rm
(8
)
.
The
s
uffici
ent
descen
t
pro
per
ty
f
or
the
conjuga
te
d
gradien
t
al
gorithm
is exp
ress
ed
as:
+
1
+
1
≤
−
‖
+
1
‖
2
f
or
≥
0
an
d
>
0
(9)
Theorem
(
1)
The
sea
rch
dir
ect
ion
(
8)
with
the
co
njugati
on
coe
ff
ic
ie
nt
and
t
he
val
ue
of
giv
e
n
by
(7)
will
achiev
e
(9) fo
r
al
l
≥
1
valu
es.
Pro
of:
by m
ath
em
atical
ind
uc
ti
on
a)
Wh
e
n
=
0
, th
e
n
0
=
−
0
→
0
0
=
−
‖
0
‖
2
<
0
b)
Assum
e that t
he
relat
ion
<
0
is t
ru
e
for
eac
h
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A n
ew
par
am
et
er in t
hr
ee
-
te
r
m
c
on
j
ugate
grad
ie
nt a
l
gorit
hm
s fo
r
un
c
onst
ra
ine
d…
(
Ala
a S
aad Ah
me
d
)
341
c)
We
pr
ov
e
t
hat
the
relat
ion
(
9
)
is
cor
r
ect
w
he
n
=
+
1
by
m
ulti
plyin
g
both
sid
es
of
t
he
(
8)
by
+
1
.
We
get:
+
1
+
1
=
−
+
1
+
1
+
+
1
−
+
1
+
1
+
1
+
1
=
−
+
1
+
1
(
1
+
)
+
+
1
If
>
0
then
+
1
+
1
<
−
+
1
+
1
(
1
+
)
+
+
1
+
1
+
1
<
0
Th
us
,
t
he reg
re
ssion
pro
pe
rty
o
f
the
ne
w
m
eth
od im
pr
ove
d
i
s prove
d.
4.
CONVE
RGE
NC
E
OF T
HE NEW
IM
PR
OVED AL
GO
RITH
M
In
t
his
sect
ion,
we
will
show
that
the
three
-
t
erm
CG
m
e
thod
with
t
he
coe
f
fici
ent
of
c
on
jug
at
io
n
and
t
he
value
of
giv
e
n
by
(
7)
is
a
bsolutel
y
converge
nt.
We
ne
ed
t
he
f
ollow
i
ng
a
ssum
pt
ion
s
to
stu
dy
the
conve
rg
e
nce
of the
new pr
opose
d
al
gorithm
:
Assu
m
pt
i
on
s
(
A1)
[10]
, [1
2]
-
[14]
We
will
i
m
po
s
e the
fo
ll
owin
g assum
ption
s
on th
e
cod
om
ain
(targ
et
)
f
unct
ion
:
a)
Level set
=
{
∈
:
(
)
≤
(
∘
)
}
is a
cl
os
ed
and
rest
rict
ed
at
the
ini
ti
al
p
oin
t.
b)
The
co
dom
ai
n
(targ
et
)
f
un
ct
io
n
is
c
on
ti
nu
ous
a
nd
der
i
vab
l
e
in
s
om
e
proxim
i
ty
of
of
le
vel
set
,
a
nd
it
s g
ra
des
a
re c
on
ti
nu
ous (
li
ps
chitz
conti
nuous
)
. T
his m
eans th
at
the
re is a
constant
>
0
, as
t
hat:
‖
(
)
−
(
)
‖
≤
‖
−
‖
∀
x,y
∈
c)
The
co
dom
ai
n
(tar
get)
f
unct
ion
is
unif
orm
ly
con
ve
x
f
un
ct
io
n
,
the
re
is
a
c
onsta
nt
nu
m
ber
tha
t
achieves
v
a
ria
nce,
a
s that:
(
(
)
−
(
)
)
(
−
)
≥
‖
−
‖
2
, for
any
,
∈
On the
oth
e
r h
and,
us
in
g
as
sum
pt
ion
s
(A1
)
t
her
e
is a
posit
ive c
onsta
nt
, as
that:
‖
‖
≤
,
∀
∈
≤
‖
(
)
‖
≤
,
∀
∈
(10)
Le
mma
[10],
[
15
]
,
[
16
]
We
prese
nt
assum
ption
s
(
A1)
and
(10)
a
re
a
chieve,
a
nd
by
ref
er
rin
g
to
(8)
f
or
the
c
onju
gate
gr
a
die
nt
wh
e
re
i
s a slopin
g se
arch directi
on, a
nd the le
ng
t
h of ste
p
is
ob
ta
i
ned f
ro
m
the st
ron
g
sea
rch li
ne
for
Wo
l
fe
.
I
f
∑
1
‖
+
1
‖
2
=
∞
>
1
we get
→
∞
(
‖
‖
)
=
0
Theorem
:
We
pro
pose
assu
m
ption
s
(A1
)
and
(10)
are
accom
plished
by
reg
res
sio
n
conditi
on
.
T
he
conjugate
gr
adient
m
et
ho
d
with
t
he
c
oeffici
ent
of
co
njugati
on
and
the
value
of
is
giv
e
n
by
(7),
as
if
is
fu
lfil
le
d
wit
h
two
str
ong
w
olf
co
ndit
ion
s
(5)
an
d
(
6).
Si
nc
e
the
co
do
m
ai
n
(tar
get)
fun
ct
ion
is
un
if
orm
l
y
convex
at
t
he
plane
of the set
, the
n t
he
e
qu
at
i
on
→
∞
‖
‖
=
0
is achie
ved.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
23
, N
o.
1
,
Ju
ly
2021
:
3
3
8
-
3
4
4
342
Pro
of:
‖
+
1
‖
=
‖
−
+
1
+
−
+
1
‖
‖
+
1
‖
≤
‖
+
1
‖
+
‖
‖
+
‖
+
1
‖
‖
+
1
‖
≤
‖
+
1
‖
(
1
+
)
+
‖
+
1
‖
2
‖
‖
2
‖
‖
‖
+
1
‖
≤
(
(
1
+
)
+
‖
+
1
‖
‖
‖
2
‖
‖
)
‖
+
1
‖
∑
1
‖
+
1
‖
≥
1
≥
(
1
(
(
1
+
)
+
‖
+
1
‖
‖
‖
2
‖
‖
)
2
)
1
2
∑
1
=
∞
us
in
g
t
he
le
m
m
a above
→
∞
‖
‖
=
0
5.
NUMER
IC
A
L RES
ULTS
In
this
sect
io
n,
we
discu
ss
th
e
nu
m
erical
resu
lt
s
of
th
e
ne
w
im
pr
ov
e
d
al
gorithm
that
we
ob
ta
ine
d
from
us
ing
t
he
ne
w
f
orm
ula
i
n
(
8)
f
or
a
set
of
te
st
functi
ons
in
un
restrict
ed
non
-
li
near
op
ti
m
iz
ation
[
17
]
.
T
o
evaluate
th
e
pe
rfor
m
ance
of
t
his
pr
opos
e
d
a
lgorit
hm
,
the
r
esults
of
(75)
f
un
ct
io
ns
[
18
]
t
hat
we
re
i
nclu
ded
in
this
stu
dy
wer
e
ch
os
e
n
t
o
c
ompare
with
the
ot
her
cl
assic
al
c
onjug
at
e
gradi
ent
m
et
ho
d
(FR
,
TTFR
),
sho
wn
i
n
the
source
[
17]
.
All
cod
es
w
ere
w
ritt
en
us
i
ng
Fortra
n
77
and
M
ATL
AB
R200
9b.
Usi
ng
a
com
par
iso
n
of
Do
la
n
an
d
Mo
re´
we
no
ti
ce
thr
ough
Fig
ur
e
s
1
-
3
a
cl
ear
su
pe
rio
rity
of
the
ne
w
i
m
pr
oved
al
go
rithm
with
resp
ect
t
o
the
nu
m
ber
of
it
er
at
ion
s
i
n
Fig
ur
e
1
a
nd
the
nu
m
ber
of
ti
m
es
the
f
unct
ion
is
ca
lc
ul
at
ed
in
F
igure
2
and
al
s
o
in
t
erm
s
of
the
c
pu
ti
m
e
ta
ken
to
i
m
ple
m
ent
the
pro
gr
am
in
Fig
ur
e
2
in
Dim
ension
s
=
100
,
200
,
…
,
1000
[
19
]
.
We
al
s
o
w
ro
te
a
Table
1
f
or
(22)
unr
est
rict
ed
non
-
li
near
opti
m
iz
a
tio
n
f
unct
io
ns
t
o
sh
ow
the
e
ff
ic
i
ency
of
t
he
ne
w
im
pr
ov
e
d
m
et
hod
f
or
num
ber
s
of
it
erati
ons
(
Iter),
an
d
t
he
num
ber
of
f
un
ct
io
n
evaluati
ons
(FE)
in
Dim
ensio
ns
100,
with
st
op
te
st
‖
+
1
‖
<
10
−
6
There
are
oth
e
r
re
sea
rch
in
t
he
sam
e
fiel
d
bu
t
with
dif
fere
nt test
fun
ct
i
ons. Fo
r
m
or
e s
ee
[
5],
[
20]
-
[
29]
.
Table
1.
U
nr
es
tric
te
d
no
n
-
li
ne
ar
op
ti
m
iz
a
ti
on
fun
ct
io
ns
t
o
s
how
the e
ff
ic
ie
nc
y of t
he ne
w
i
m
pr
ov
e
d
m
et
ho
d
for nu
m
ber
s
of
it
erati
on
s
(I
te
r
),
a
nd the
num
ber o
f funct
io
n evaluat
i
on
s
(
F
E)
in
D
im
ensio
ns
100
Prob
le
m
s
Di
m
Iter
FE
NE
W
FR
TT
FR
NE
W
FR
TT
FR
Freud
en
stein
&
Ro
th
100
84
1529
328
1979
4
4
0
9
2
9131
Exten
d
ed
Ro
sen
b
r
o
ck
100
35
43
42
73
87
82
Exten
d
ed
W
h
ite
&
Hols
t
100
32
37
35
68
77
70
Exten
d
ed
Beale
B
EAL
E
100
14
15
15
27
29
28
Perturb
ed
Quad
rati
c
100
96
101
100
144
155
153
Raydan
1
100
78
90
86
118
138
133
Diag
o
n
al 2
100
63
64
71
105
105
121
Diag
o
n
al 3
100
190
203
242
2772
3075
4869
Hag
er
100
27
47
31
44
565
49
Gen
eralize
d
Tr
id
ia
g
o
n
al 1
100
23
23
24
45
45
50
Exten
d
ed
Powell
100
53
71
79
101
136
151
Exten
d
ed
Clif
f
100
12
f
ail
19
30
f
ail
46
Qu
ad
ratic
Diag
o
n
a
l
100
53
54
53
95
95
96
Exten
d
ed
Hiebert
100
78
87
85
170
188
187
Exten
d
ed
Quad
ratic
Pen
alty
100
24
29
25
55
61
53
BDQR
TI
C
100
310
532
587
6084
1
1
6
6
4
1
1
4
5
7
TRIDI
A
100
348
364
392
542
566
624
Bro
y
d
en
T
ridiag
o
n
al
100
29
31
31
53
49
49
Tr
id
iag
o
n
al Per
tu
r
b
ed
Quad
ratic
100
99
109
105
157
173
167
Exten
d
ed
DE
NSC
HNC
100
14
15
18
26
27
31
BIGGSB1
100
477
617
533
750
985
837
Exten
d
ed
Blo
ck
-
Diag
o
n
al
100
14
15
15
24
26
26
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
A n
ew
par
am
et
er in t
hr
ee
-
te
r
m
c
on
j
ugate
grad
ie
nt a
l
gorit
hm
s fo
r
un
c
onst
ra
ine
d…
(
Ala
a S
aad Ah
me
d
)
343
Figure
1
.
A co
m
par
ison
of th
e num
ber
of ite
rati
on
s
Figure
2
.
A co
m
par
ison
of th
e num
ber
of ti
m
es a f
unct
io
n i
s
cal
culat
ed
Figure
3
.
A co
m
par
ison
of th
e tim
e taken
to
p
e
rfor
m
the fu
nctions
6.
CONCL
US
I
O
NS
We
pr
es
ente
d
i
n
this
resea
rc
h
a
ne
w
ty
pe
of
TTCG
al
gorith
m
to
so
l
ve
t
he
pro
blem
s
of
un
const
raine
d
op
ti
m
iz
ation
,
a
nd
t
he
propos
ed
al
gorithm
ha
s
show
n
a
hi
gh
e
f
fici
e
ncy
in
s
olv
in
g
the
s
e
pro
blem
s
wi
th
the
le
ast
n
um
ber
of it
erati
ons a
nd w
it
h hig
he
r
ac
cur
acy
i
n
re
ach
ing
t
he
a
ppr
ox
i
m
at
e so
luti
on
of the
functi
on.
ACKN
OWLE
DGE
MENTS
I
w
ould
li
ke
to
ex
pr
ess
m
y
si
ncer
e
grat
it
ud
e
an
d
ap
preci
at
ion
to
m
y
su
perviso
rs
P
r
of
.
D
r
.
K
halil
K.
Abb
o
f
or this
val
uab
le
s
uggest
ion
,
en
c
oura
ge
m
ent an
d i
nval
uab
le
rem
ark
duri
ng writ
ing
t
his
pap
e
r.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
23
, N
o.
1
,
Ju
ly
2021
:
3
3
8
-
3
4
4
344
REFERE
NCE
S
[1]
R.
Flet
ch
er
and
C.
M.
Ree
ves,
“
Functi
on
m
ini
m
iz
ation
b
y
con
jug
at
e
gra
d
ie
nts
,
”
C
omput
.
J.
,
vol.
7
,
no.
2,
pp.
149
-
154,
1964
,
doi
:
1
0.
1093/c
om
jnl/7
.
2.
149
.
[2]
Y.
H.
Dai
and
Y.
Yuan,
“
A
Nonlinear
Conjugat
e
Gradi
en
t
Meth
od
with
a
Stron
g
Global
Conve
rge
nce
Prope
rt
y
,
”
SIAM
J
.
Opt
im.
,
vol.
10
,
no
.
1
,
pp
.
177
-
182
,
1999
,
doi: 10.
1137
/S105262349731
8992.
[3]
H.
M.
Khudhur
and
K.
K.
Abbo
,
“
A
New
Conju
gat
e
Gradi
ent
Method
for
Le
arn
i
ng
Fuzz
y
Neura
l
Networks,”
J.
Mult
idi
sc
ip. Mo
del
.
Optim.
,
vo
l.
3,
no
.
2
,
pp
.
57
-
69,
2021
.
[4]
A.
S.
Ahm
ed, “O
pti
m
iz
at
io
n
M
et
hods For Lear
ning
Artif
ic
i
al Neura
l
Networks,
”
Un
ive
rsi
t
y
of
Mos
ul,
2018.
[5]
Hisham
M.
Kh
udhur;
Khali
l
K.
Abbo,
“
A
New
T
y
pe
of
Conjuga
te
Grad
ie
nt
T
ec
hniqu
e
for
Solving
Fuz
z
y
Nonline
ar
Algeb
rai
c
Equ
at
ions,
” 2020.
[6]
M.
R.
Heste
nes
and
E.
Stie
f
el,
"
Methods
of
conj
ugat
e
gra
d
ie
nts
for
solving
li
near
sy
st
em
s
,
"
vol.
49,
no.
1.
NBS
W
ashingt
on,
DC
,
1952
.
[7]
L.
C.
W
.
Dixon
,
“
Conjuga
te
gr
adi
en
t
al
gori
thms
:
quadr
at
i
c
te
r
m
ina
ti
on
withou
t
li
ne
ar
sea
rch
es,
”
IMA
J
.
Appl.
Math.
,
vol
.
15
,
n
o.
1
,
pp
.
9
-
18
,
1
975.
[8]
E.
Polak
and
G
.
Ribi
er
e,
“
Note
sur
la
conv
erg
en
ce
d
e
m
ét
hodes
de
direct
ions
co
njugué
es,
”
ES
AI
M
Math.
Mod
el.
Numer.
Anal.
M
athé
matiqu
e
Ana
l.
Numériqu
e
,
vo
l.
3
,
no
.
R1
,
pp
.
35
-
43,
1969
.
[9]
L.
Zha
ng
and
W
.
Zhou,
“
Two
desc
ent
h
y
b
rid
conj
ugate
gra
di
ent
m
et
hods
for
opti
m
iz
at
ion
,
”
J.
Comput.
App
l.
Math.
,
vol
.
216
,
no.
1
,
pp
.
251
-
2
64,
2008
.
[10]
K.
K.
ABBO
,
Y.
A.
Lay
l
ani,
and
H.
M.
K
hudhur,
“
A
New
Spect
ral
Co
njuga
t
e
Gradi
en
t
Algorit
hm
For
Unconstra
ine
d
Optimiza
ti
o
n,
”
Int.
J
.
Math.
C
omput.
Appl.
R
es.
,
vol
.
8,
pp.
1
-
9,
2018,
[Online
]
.
Availabl
e
:
ww
w.t
jprc
.
org
.
[11]
H.
N.
Jabba
r
,
K
.
K.
Abbo,
and
H.
M.
Khudhur,
“
Four
--
Te
rm
Conjuga
t
e
Gradi
e
nt
(CG)
Method
Based
on
Pure
Conjuga
c
y
Cond
it
ion for Unc
ons
tra
in
ed
Optimiz
a
ti
on,
”
ki
rkuk
Uni
v.
J. Sci. Stud.
,
v
ol.
13
,
no
.
2
,
pp
.
10
1
-
113,
2018
.
[12]
H.
M.
Khudhur
and
K.
K.
Abb
o,
“
New
h
y
brid
of
Conjuga
t
e
Gradi
ent
Techni
que
for
Solving
Fuzz
y
Nonlinear
Equa
ti
ons
,
”
J. S
oft
Comput
.
Arti
f
.
Int
el
l
.
,
vol
.
2
,
n
o.
1
,
pp
.
1
-
8
,
20
21.
[13]
K.
K.
Abbo
and
H.
M.
Khudhur,
“
New
A
hy
b
rid
Heste
nes
-
Stie
f
el
and
Dai
-
Yuan
c
onjuga
t
e
gra
di
en
t
al
gori
thms
fo
r
unconstra
in
ed
o
pti
m
iz
ation,
”
Ti
k
rit
J
.
Pure
S
ci
.
,
v
ol.
21
,
no
.
1
,
pp
.
118
-
123,
2016
.
[14]
K.
K.
Abbo
and
H.
M.
Khudhur,
“
New
A
hy
bri
d
conj
ugate
gra
d
ie
nt
Fle
tc
her
-
Re
eve
s
and
Polak
-
Ribi
er
e
al
gori
th
m
for
unco
nstra
ined opt
imiz
at
ion
,
”
Tikrit
J. P
ure
S
c
i.
,
vol
.
21
,
no
.
1
,
pp.
124
-
129
,
20
16.
[15]
K.
K.
Abbo,
Y
.
A.
Lay
l
ani,
a
nd
H.
M.
Khu
dhur,
“
Propos
ed
new
Scaled
c
onjuga
t
e
gra
d
ient
al
gor
it
hm
fo
r
Unconstra
ine
d
Optimiza
t
ion,”
In
t
.
J
.
Enhanc. Re
s.
Sci.
Te
chnol.
En
g.
,
vol. 5, no. 7,
2016.
[16]
Z.
M.
Abdul
la
h
,
M.
Ham
ee
d,
M
.
K.
Hisham
,
an
d
M.
A.
Khal
ee
l
,
“
Modifie
d
new
conj
uga
te
gr
adient
m
et
hod
fo
r
Unconstra
ine
d
Optimiza
t
ion,”
Ti
k
rit
J
.
Pure
S
ci
.
,
v
ol.
24
,
no
.
5
,
pp
.
86
-
90,
2019
.
[17]
N.
Andrei
,
“
An
unconstra
in
ed
o
pti
m
iz
at
ion
t
est
func
ti
ons
co
ll
e
ction,
”
Adv.
Mode
l.
Optim
,
vol
.
10
,
no.
1
,
pp
.
147
–
161,
2008
.
[18]
H.
M.
Khudhur,
“
Nu
m
eri
ca
l
and
ana
l
y
t
ical
stud
y
of
som
e
desc
ent
al
gor
it
hm
s
to
solve
unconstr
ai
n
ed
Optimization
proble
m
s,”
Univ
ersity
of
Mos
ul,
2015.
[19]
E.
D.
Dolan
an
d
J.
J.
Moré,
“
B
enc
hm
ark
ing
op
ti
m
iz
ation
software
with
per
for
m
anc
e
profil
es
,
”
Math.
Program
.
,
vol.
91
,
no
.
2
,
pp
.
201
-
213
,
2002
.
[20]
B.
A.
Hass
an,
H.
O.
Daha
wi,
an
d
A.
S.
Younus
,
“
A
new
kind
of
par
amete
r
con
j
ugat
e
gra
d
ie
nt
f
or
unconstra
ine
d
opti
m
iz
ation,
”
I
ndones.
J
.
Elec
tr
.
Eng
.
Comput
.
S
ci
.
,
vo
l. 17, no. 1
,
2019
,
doi
:
10
.
1
1591/i
jeec
s.v17
.
i1.
pp404
-
411.
[21]
B.
A.
Hass
an
an
d
H.
K.
Khalo
,
“
A
new
c
la
ss
of
BF
GS
updat
ing
form
ula
base
d
o
n
the
new
qu
asi
-
newton
equ
at
ion
,
”
Indone
s. J.
Elec
t
r.
Eng
.
Co
mput.
Sci
.
,
vo
l. 13, no.
3,
2019
,
doi
:
10
.
11591/i
jeec
s.v13
.
i3.
pp945
-
953
.
[22]
B.
A.
Hass
an,
“A
new
t
y
pe
of
quasi
-
newton
upd
at
ing
form
ula
s
base
d
on
the
new
quasi
-
newton
e
quat
ion
,
”
Numer
.
Al
gebr.
Control
Optim.
,
vol. 10,
no.
2
,
2020
,
doi
: 10.3934/
na
co.
20
1904
9.
[23]
B.
A.
Hass
an,
Z.
M.
Abdull
ah,
and
H
.
N.
Jab
bar
,
“
A
desc
ent
extension
of
th
e
Dai
-
Yuan
c
onjuga
t
e
gra
d
ient
te
chn
ique
,
”
Indo
nes.
J
.
Elec
tr.
E
ng.
Comput.
Sc
i.
,
vol
.
16
,
no
.
2
,
2
019,
doi
:
10
.
115
91/i
jeec
s.v16
.
i2
.
pp661
-
668.
[24]
B.
A.
Hass
an,
A.
O.
Ow
ai
d,
and
Z.
T
.
Yasen,
“
A
var
ia
n
t
of
h
y
brid
conj
uga
te
gr
adient
m
et
hods
bas
ed
on
the
convex
combinat
ion
fo
r
opti
m
iz
ation,
”
Indone
s.
J.
El
ectr.
Eng.
Comput.
Sci
.
,
vol.
20,
no
.
2,
2020,
d
oi
:
10.
11591/ijeecs.
v20.
i2.
pp1007
-
1
015.
[25]
B.
A.
H
assan
and
M.
W
.
T
a
ha,
“
A
m
odifi
e
d
quasi
-
Newton
equation
in
t
he
quasi
-
Newto
n
m
et
hods
for
opti
m
iz
ation,
”
A
ppl.
Ma
th. Sc
i
.
,
vol.
13
,
no
.
10
,
2
019,
doi
:
10
.
129
88/a
m
s.2019.
93
51.
[26]
B.
A.
Hass
an,
“
A
new
form
ula
for
conj
uga
te
p
a
ramet
er
computat
ion
b
ase
d
on
t
he
quadr
a
ti
c
m
odel
,
”
Indone
s.
J.
El
e
ct
r.
Eng. Co
mput.
Sc
i.
,
vo
l. 1
3,
no
.
3
,
2019
,
d
oi:
10
.
11591/ije
e
cs.
v13.
i3
.
pp954
-
961.
[27]
B.
A.
H
assan
a
nd
M.
W
.
Ta
ha
,
“
A
new
var
i
ants
of
quasi
-
newt
on
equation
b
ase
d
on
the
qu
adr
at
i
c
func
ti
on
fo
r
unconstra
in
ed
opti
m
iz
ation
,
”
Indone
s.
J
.
E
le
c
tr.
Eng
.
C
omput.
Sc
i.
,
vol.
19
,
no
.
2,
2020
,
d
oi:
10.
11591/ijeecs.
v19.
i2.
pp701
-
70
8.
[28]
H.
M.
Khudhur,
“
Modifie
d
Barzi
la
i
-
Borwe
in
Met
hod
for
Stee
p
est
Desce
nt
Me
tho
d
to
Solving
Fuz
z
y
Optimiz
at
ion
Problems
(FO
P),
”
A
lbahi
r
J.
,
vol
.
12,
no.
23
-
24,
p
p.
63
-
72
,
2020
.
[29]
K.
K
Abbo
and
F.
H
Mohamm
ed,
“
Spect
ral
Fl
et
ch
er
-
Re
eve
s
Algorit
hm
for
Sol
ving
Non
-
Li
near
Unconstra
ine
d
Optimiza
ti
o
n
Pr
oble
m
s,”
Iraqi
J
.
Stat.
S
ci
.
,
vo
l. 1
1,
no
.
19
,
pp
.
21
-
38,
2011
.
Evaluation Warning : The document was created with Spire.PDF for Python.