Int
ern
at
i
onal
Journ
al of
P
ower E
le
ctr
on
i
cs a
n
d
Drive
S
ystem
s
(
IJ
PEDS
)
Vo
l.
12
,
No.
2
,
Jun
2021
,
pp.
119
6
~
120
4
IS
S
N:
20
88
-
8694
,
DOI: 10
.11
591/
ij
peds
.
v12.i
2
.
pp
119
6
-
120
4
1196
Journ
al h
om
e
page
:
http:
//
ij
pe
ds
.i
aescore.c
om
Improvi
ng neura
l netwo
rk
us
i
ng a sine tr
ee
-
seed al
gorithm
f
or
tuning m
otor D
C
Widi
A
ri
b
owo
, Ba
m
bang
Supri
anto,
Joko
Depa
rtment
o
f
E
le
c
tri
c
al E
ngin
eering,
Univ
ersitas Nege
ri
Suraba
y
a,
Indon
esia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
hist
or
y:
Re
cei
ved
Oct
6
, 2
0
20
Re
vised
Feb
2,
20
21
Accepte
d
M
ar
2
3
, 20
21
A
DC
mot
or
is
appl
i
ed
to
de
li
c
at
e
spee
d
and
p
ositi
on
in
th
e
in
dustry.
Th
e
stabi
lity
and
pr
oduct
ivity
of
a
sys
te
m
are
keys
for
tuni
ng
of
a
DC
mot
or
spee
d.
Stab
il
i
zed
spee
d
is
influence
d
by
loa
d
sw
ay
and
env
ironm
ental
fac
tors.
In
thi
s
pape
r,
a
com
par
ison
stu
dy
i
n
di
ver
se
t
ec
hn
ique
s
to
tune
th
e
spee
d
of
the
DC
mo
tor
wi
th
par
a
me
t
er
un
ce
r
ta
in
t
ie
s
is
show
ed.
T
he
r
ese
ar
c
h
has
discussed
t
he
app
li
c
ation
o
f
the
fe
ed
-
forw
ard
neu
ral
ne
twork
(FF
NN
)
which
is
enha
n
c
ed
by
a
sin
e
tr
e
e
-
see
d
al
gor
it
hm
(STSA
).
STS
A
is
a
hybrid
me
thod
of
the
t
ree
-
see
d
al
gori
t
hm
(TSA)
and
Sine
Cosine
a
lg
orit
hm
.
The
STS
A
me
thod
i
s
ai
m
ed
to
im
p
rove
TSA
per
fo
rma
nc
e
b
ase
d
o
n
the
sin
e
cosine
a
lgori
th
m
(SCA
)
me
thod.
A
fee
d
-
forward
neur
al
n
et
work
(FF
NN
)
i
s
popula
r
and
ca
p
abl
e
of
nonl
inea
r
issues.
Th
e
fo
c
us
of
the
rese
arch
is
on
the
ac
hi
eve
m
ent
spe
ed
of
DC
mot
or
.
In
addi
t
ion,
th
e
proposed
me
t
hod
will
b
e
com
par
ed
with
proporti
ona
l
i
nte
gra
l
der
iva
t
i
ve
(PID
),
FF
NN
,
ma
r
ine
pre
dat
or
a
lgori
th
m
-
fee
d
-
forw
ard
neur
al
ne
twork
(
MP
A
-
N
N)
and
a
tom
se
arc
h
al
gorit
h
m
-
fe
ed
-
f
orward
neur
al
n
et
wor
k
(AS
O
-
N
N).
The
p
erf
orm
anc
e
of
the
spee
d
from
the
proposed
me
tho
d
has
the
b
est
r
e
sult.
Th
e
set
tl
in
g
ti
m
e
va
lue
of
the
proposed
me
thod
is
mor
e
stabl
e
tha
n
th
e
P
ID
me
thod
.
The
ITAE
va
lue
of
the
STS
A
-
NN
me
thod
was
31.
3%
bet
t
er
th
an
t
he
PI
D
me
thod
.
Mea
nwhile,
the
I
TSE
v
al
ue
i
s 29.
2%
b
et
t
er th
an
th
e
PID
m
et
h
od
.
Ke
yw
or
d
s
:
DC m
otor
Feed
-
f
orward
neural
netw
ork
Sine c
os
ine
alg
or
it
hm
Sine tree
-
see
d al
gorithm
Tree
-
see
d
al
go
rithm
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
BY
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Widi
Ar
i
bow
o
Dep
a
rteme
nt of Elect
rical
E
nginee
rin
g
Un
i
ver
sit
as
Ne
ger
i
Suraba
ya
Un
es
a Ka
m
pu
s
Keti
ntan
g, S
urabay
a
61
256, J
awa Tim
ur,
Indonesia
Emai
l:
w
idiari
bow
o@u
nesa.a
c.id
1.
INTROD
U
CTION
DC
mo
t
or
(
di
rect
c
urren
t
)
i
s
a
basic
el
ec
trom
ec
ha
nical
dev
ic
e
that
ha
s
the
f
unct
io
n
to
co
nvert
el
ect
rical
po
we
r
into
mecha
ni
cal
power.
DC
mo
to
r
is
a
ty
pe
of
m
ot
or
that
is
us
ed
direct
vo
lt
age
a
s
it
s
powe
r
so
urce.
By
pro
vid
in
g
a
volt
ag
e
dif
fer
e
nce
at
the
tw
o
te
r
min
al
s,
the
mo
t
or
will
be
ro
ta
ti
ng
in
one
direct
ion
.
I
f
the
po
la
rit
y
of
the
volt
age
is
re
ver
se
d,
the
directi
on
ro
ta
ti
on
of
the
mo
t
or
will
be
reversin
g
as
well
.
Th
e
po
la
rity
of
the
vo
lt
age
is
a
ppli
ed
in
the
t
wo
t
erminals
to
det
ermine
the
dire
ct
ion
of
r
otati
on
of
the
m
oto
r
.
DC
mo
to
rs
ha
ve
re
li
abili
ty,
flexib
il
it
y,
and
lo
w
cost.
It
is
the
r
easo
n
w
hy
DC
mo
t
or
is
ve
r
y
popula
r
i
n
in
dustria
l
app
li
cat
io
ns
an
d
ho
us
e
ho
l
d
a
pp
li
ance
s
that
requires
mo
t
or
sp
ee
d
an
d
posit
ion
co
ntr
ol.
DC
m
oto
r
s
are
mo
s
t
com
patible
with level
sp
ee
d
c
on
t
ro
l a
nd are
t
her
e
fore a
ppli
ed
in
ma
ny ada
pt
able sp
ee
d dr
i
ves [1].
DC
m
otors
ha
ve
bette
r
spe
e
d
c
har
act
erist
i
cs
than
AC
m
otors.
In
ad
diti
on,
t
he
DC
m
otors
hav
e
a
n
op
ti
mal
sp
ee
d
con
t
ro
l
f
or
br
e
akin
g
a
nd
acc
el
erati
on
.
DC
mo
to
rs
ha
ve
a
lo
ng
e
r
li
feti
me
due
to
a
dju
st
ments
and v
a
riat
ions
in imple
mentat
ion
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
P
ow Elec
& Dri S
ys
t
IS
S
N: 20
88
-
8
694
Impr
ovin
g neural
network
us
i
ng a sine
tree
-
s
eed
algorit
hm f
or
t
un
i
ng moto
r D
C
(
Wi
di
Aribo
w
o
)
1197
Conve
ntion
al
mo
to
r
co
ntr
ollers
su
c
h
as
P
I
D
co
ntr
ollers
a
re
sti
ll
po
pula
r
ly
ap
plied
[
2].
This
is
due
to
the
sim
ple
P
I
D
t
opolog
y.
T
he
wea
kn
es
s
of
P
I
D
is
t
o
ac
hieve
hi
gh
performa
nce
from
a
c
ontrolle
r
t
hat
has
require
d
accu
r
at
e
an
d
preci
se
co
ntr
ol
pa
ram
et
ers.
A
good
PI
D
c
ontrol
set
ti
ng
will
hav
e
an
im
pact
on
opti
mal
sy
ste
m
res
ponse
[
3]
.
It
is
rea
ll
y
de
p
en
ding
on
the
par
a
mete
r
set
ti
ng
s
an
d
the
mat
hemati
cal
model.
Cl
a
ssica
l
methods
ha
ve
char
act
e
risti
cs
wh
e
n
modeli
ng
a
non
-
li
near
sy
ste
m.
T
his
is
infl
uen
ce
d
by
the
co
ntr
ol
e
quat
ion
wh
ic
h
is c
ompl
ic
at
ed
an
d req
ui
red
more e
ffort
.
This
is
sti
m
ulate
d
t
he
de
velopm
ent
of
int
el
li
gen
t
c
on
t
rol
te
ch
no
l
ogy
[
4].
T
he
dev
el
opment
of
arti
fici
al
intel
li
gen
ce
is
becomi
ng
a
n
a
dd
it
ion
al
s
upplem
ent
in
te
rms
of
c
on
t
ro
l.
I
n
r
ecent
year
s
,
s
ever
al
researc
hers
ha
ve
sta
rted
t
o
a
pp
l
y
se
ve
ral
in
te
ll
igent
co
ntr
ol
met
hods
ba
sed
on
a
rtific
i
al
intel
li
gen
ce.
Neural
netw
orks
a
re
the
fa
vorite
m
et
hod
us
ed
be
cause
the
y
ha
ve
the
a
bili
ty
to
s
olv
e
c
omplex
a
nd
no
n
-
li
near
pro
blems.
Se
ve
ral
neural
net
work
meth
od
s
are
a
ppli
ed
in
app
l
ic
at
io
ns
suc
h
a
s
f
or
ecast
i
ng
[
4
],
cl
assifi
cat
ion
[
5
],
est
imat
ion
[
6
]
,
a
n
d
pr
e
dicti
on
[
7
].
Se
ve
ral
stu
dies
on
DC
m
otor
c
on
trol
a
re
div
i
de
d
i
nto
t
wo
c
oncepts,
namely:
c
onve
ntion
al
meth
ods
an
d
arti
fici
al
intel
li
gen
ce
m
et
hods
.
seve
ral
arti
fici
al
intel
l
igence
c
oncept
s
f
or
con
t
ro
ll
in
g
s
uc
h
as:
W
hale
opti
miza
ti
on
al
go
rithm
[
8
],
[
9
]
,
Harris
Hawks
op
ti
miza
ti
on
al
gorithm
[
10
],
f
lowe
r
po
ll
inati
on
al
gorith
m
[
11
],
fi
ref
ly
al
gorithm
[
12
]
,
ant
c
olony
al
gorithm
[
13
],
[
14
],
a
nd
neural
net
wor
k
[
15
],
[
16
].
Feed
for
ward
neural
net
wor
k
(F
F
N
N)
has
a
simple
an
d
ea
s
y
to
im
pleme
nt
netw
ork
st
ru
ct
ur
e
.
FF
N
N
is
the
fa
vorite
of
t
he
m
os
t
widely
imple
mented
an
d
is
gro
wing
ra
pi
dly
with
a
wide
va
riet
y.
Se
ver
al
com
pu
ta
ti
on
methods
ha
ve
bee
n
wi
dely
app
li
ed
to
op
t
imi
ze
neural
ne
tworks
su
c
h
as
the
par
ti
cl
e
swar
m
op
ti
miza
ti
on
(
PSO)
meth
od
wh
ic
h
is
a
pp
li
ed
f
o
r
wei
gh
ti
ng
F
FNN
opti
miza
ti
on
[
17
],
Ge
netic
al
gor
it
hm
com
bin
e
d wit
h FF
NN [
18
], a
nd a c
ombinati
on of
gr
a
vitat
ion
al
searc
h
al
g
or
it
hm a
nd FFNN
[
19
]
.
Tree
-
see
d
al
gorithm
(TSA)
ha
s
the
a
dv
a
nta
ge
that
it
can
s
olv
e
opti
miza
tio
n
pro
blems
.
On
t
he
oth
e
r
hand,
TSA
has
li
mit
ed
exp
l
oitat
ion
w
hen
de
al
ing
with
c
omplex
pro
blems.
The
sine
-
c
os
ine
al
gorith
m
(
SCA
)
method
has
th
e
a
dv
a
ntage
of
ba
la
ncin
g
ex
pl
or
at
io
n
a
nd
e
xp
l
oitat
ion
.
T
he
SC
A
c
an
fin
d
pro
misi
ng
a
r
eas
of
the searc
h spac
e an
d ult
imat
el
y
c
onve
rg
e
to
t
he glo
bal opti
mal [
20
].
T
his
resea
rc
h
is
proposi
ng
t
he
sine
t
ree
-
se
ed
al
gorith
m
method
t
o
imp
rove
neural
ne
twork
s
kill
s.
The
li
mit
at
ions
of
the
ne
ur
al
netw
ork
i
n
w
ei
gh
ti
ng
will
be
ha
nd
le
d
by
the
sine
tree
-
se
ed
al
go
rithm
method.
The
meth
od
of
the
sine
tree
-
s
eed
al
gorithm
i
s
de
vel
op
e
d
by
co
mb
i
ning
t
he
sine
co
sine
al
gorithm
a
nd
th
e
tree
seed
al
go
rithm
[
20
]
.
T
he
met
hod
of
t
his
res
earch
is
cal
le
d
the
STS
A
-
N
N
hybri
d
met
hod.
The
meth
od
is
us
e
d
to
co
ntr
ol
the
sp
ee
d
of
a
DC
mo
t
or
.
The
c
on
t
rib
ution
of
this
resea
rch
is
to
prese
nt
int
el
li
gen
t
co
ntr
ol
sk
il
l
s
base
d
on
t
he
Sine
T
ree
-
Se
ed
Algo
rithm
hy
br
i
d
meth
od
a
nd
the
ne
ur
al
net
wor
k.
T
he
validat
ion
a
nd
eff
ect
ive
ness o
f
the
prop
os
e
d met
hod
a
re c
ompa
red with
th
e
PID
,
FF
NN
,
MP
A
-
N
N a
nd
AS
O
-
NN
.
2.
MA
TE
RIA
L
S
AND
METH
OD
S
2.1.
A
sine
tree
-
se
ed
al
go
ri
th
m
The
met
hod
pro
posed
by
S
TSA
-
N
N
is
a
com
bin
at
io
n
of
S
TS
A
an
d
NN
met
hods.
The
STS
A
al
gorithm
is
a
n
imp
r
ov
e
men
t
in
the
Tree
-
Seed
Algorith
m
(TSA)
met
hod
us
in
g
the
Sine
Cosi
ne
(S
CA
)
method.
2.1.1.
TSA: T
ree
-
see
d a
l
go
ri
th
m
TSA
is
a
met
aheurist
ic
met
hod
that
has
a
f
unct
ion
as
an
i
ntell
igent
op
ti
miza
ti
on
ba
sed
on
the
relat
ion
s
hip
be
tween
trees
a
nd
seeds
.
It
is
use
d
to
s
olv
e
ongoin
g
pr
ob
le
ms.
At
the
sta
r
t
of
t
he
al
gorit
hm
,
t
he
trees
are
plant
ed
a
bove
the
gro
und.
T
he
loc
at
ion
of
t
he
tre
e
is
placed
wit
h
the
best
poss
ible
so
luti
on
t
o
the
pro
blem.
The
l
and
is
a
re
pres
entat
ion
of
t
he
search
s
pace.
A
c
ontrol
pa
ra
mete
r
cal
le
d
se
arch
te
nde
ncy
(S
T
)
is
us
e
d
to
handl
e
trees
t
hat
ha
ve
t
he
best
s
earch
te
nde
n
c
y
or
ta
ke
n
ra
ndom
l
y.
The
TSA
meth
od
has
t
he
adv
a
ntage
of st
rengthe
ning a
nd
blen
ding lo
c
al
ly to op
ti
mal
or n
ea
r
-
opti
mal.
M
ea
nwhile
,
a
very
imp
or
ta
nt
opti
miza
ti
on
pro
blem
is
t
o
ge
t
a
cl
ear
l
ocati
on
gen
e
rated
from
the
tree.
This
is
t
he
f
oc
us
of
the
sea
r
ch.
N
e
w
trees
are
create
d
by
re
placi
ng
exis
ti
ng
trees
with
ne
w
see
ds
t
ha
t
are
deeme
d
s
uitab
le
and
be
st.
T
he
tree
an
d
se
ed
sea
rch
met
hod
will
be
re
peati
ng
unti
l
a
sp
eci
fie
d
nu
mb
e
r
of
funct
io
n
crit
eri
a are acc
ommo
dated. I
n
t
he
ea
rly
sta
ges, i
niti
al
iz
e the tree is
deter
mine
d by
.
,
+
1
=
,
+
,
×
(
,
−
,
min
)
(1)
,
+
1
=
,
+
,
×
(
,
−
,
)
(2)
,
+
1
=
,
+
1
+
,
×
(
,
−
,
)
(3)
,
=
min
(
∫
(
,
+
1
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
)
)
(4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
D
ri
S
ys
t,
V
ol
.
12
, N
o.
2
,
J
une
2021
:
119
6
–
120
4
1198
Wh
e
re
dime
nsi
on
of
the
tree
is
,
+
1
.
T
he
higher
boun
d
of
the
search
s
pace
is
,
.
T
he
l
ow
e
r
boun
d
of
the
searc
h
sp
a
ce
is
,
min
.
The
r
a
ndom
value
w
it
h
range
[
0,1]
gen
e
rat
ed
for
each
dime
ns
i
on
is
,
.
T
he
dimensi
on
of
the
seed
is
,
+
1
.
,
.
,
a
re
the
scal
i
ng
el
eme
nt
w
hich
gen
e
rated
in
series
of
[−1,
1]
rand
om
ly
.
The
dimens
i
on
of
top
tree
place
is
,
.
The
dime
ns
io
n
of
a
tree
randoml
y
c
hosen
f
rom
th
e
popula
ti
on
is
,
.
Fo
r
al
l
te
st,
the
sto
p
sta
te
is
the
pea
k
ti
m
es
of
functi
on
evaluati
ons
(
MaxFE
)
an
d
t
he
amo
un
t
of fu
nc
ti
on
e
valuati
on
s (
FE
)
.
=
10000
(5)
=
(6)
W
he
re
M
is t
he
d
ime
ns
io
n of
a f
un
ct
io
n,
A
is
the
numb
e
r
of
trees.
2.1.2.
SCA:
S
ine
–
c
osi
ne a
l
go
ri
t
hm
SCA is served t
o
so
lve
opti
miza
ti
on
pr
ob
le
m
s w
it
h
un
know
n
searc
h
s
pace
s u
sin
g
the si
ne
an
d
c
os
ine
functi
ons.
T
he
locat
ion
of
eac
h
sea
rch
a
ge
nt
is
updated
to
ge
t
the
opti
mal
so
luti
on.
SCA
has
a
n
al
gorith
m
f
or
com
bin
in
g
hi
gh
rand
omness
o
f
se
ver
al
s
olu
ti
ons
a
nd
r
andom
s
olu
ti
ons
w
hich
is
us
ef
ul
for
obta
ining
promisin
g
a
rea
s in
t
he
sea
rch
sp
ace
.
+
1
=
+
1
×
sin
(
2
)
×
|
3
×
1
−
|
,
4
<
0
.
5
(7)
+
1
=
+
1
×
cos
(
2
)
×
|
3
×
1
−
|
,
4
≥
0
.
5
(8)
Wh
e
re
t
he
po
si
ti
on
o
f
the
c
urr
ent
s
olu
ti
on
is
+
1
.
The
best so
l
ution
obtai
ned
s
o
far
is
1
.
1
dicta
te
s
the n
ex
t
place
(or
dire
ct
ion
of
m
ove
ment)
,
t
he
dis
ta
nce
from
w
hich
the
m
ov
e
ment
s
houl
d
move
t
o
t
he
t
arg
et
or
ou
t
ward
is
2
.
A
ra
ndom
wei
gh
t
for
t
he
go
al
s
o
th
at
stoc
hastic
al
ly
em
ph
asi
z
e
(
3
>1)
or
deem
ph
a
siz
e
(
3
<1
)
the
eff
ect
of
desali
natio
n
in
def
i
ning
the
distance
is
3
.
A
unif
orm
sw
it
ching
betwee
n
sine
a
nd
co
sin
e
functi
ons
is
4
.
=
−
(9)
Wh
e
re
is t
he c
urren
t i
te
rati
on
,
is t
he
ma
xim
um
it
erati
ons
, a
nd
is a co
ns
ta
nt.
2.1.3.
STSA
:
Sine
tr
ee
-
seed
algorit
hm
Seeds ha
ve
a
n i
mp
ort
ant
ro
le
in the dist
rib
ution an
d
sea
rch f
or
opti
mal valu
e. Th
is i
s
not o
ptimal
du
e
to
ra
ndom
a
nd
simple
see
d
pr
oductio
n.
P
oo
r
seed
producti
on
will
resu
lt
i
n
op
ti
miza
ti
on
resu
lt
s
t
hat
are
not
in
accor
da
nce
with
the
opti
mal
s
olu
ti
on.
The
S
TSA
meth
od
modifie
s
the
num
ber
of
see
ds
(
ns
)
val
ue
s
o
that
it
can
be
proces
s
ed
acc
ordin
g
to
c
ha
ng
es
in
th
e
FE
value
.
T
hi
s
has
an
ef
fect
on
t
he
a
mou
nt
that
im
pacts
on
th
e
best s
olu
ti
on
find
i
ng m
od
el
.
=
(10)
ℎ
=
0
.
5
×
×
(11)
=
×
|
(
−
)
×
cos
(
ℎ
)
|
+
1
(12)
=
2
×
(
1
−
)
(13)
Wh
e
re
th
e
nu
mb
e
r
of
see
ds
i
s
.
an
d
are
the
lo
w
a
nd upper
bo
und
of
the
num
ber
of
see
ds
pr
oduce
d
by
a
tree
.
A
new
pa
rameter
base
d
on
the
ori
gin
al
TSA
w
hich
c
ombines
the
i
nspirat
io
n
of
S
CA
is
.
has
ra
ng
e
[0,2].
TS
A
has
a
wea
kn
es
s
in
the
see
d
posit
ion
updatin
g
functi
on.
Prob
le
ms
a
rise
w
he
n
the
see
d
posit
ion
s
a
re
separ
at
e
d.
T
he
posit
io
n
of
th
is
seed
will
ha
ve
a
la
r
ge
unc
ertai
nty
val
ue.
This
will
be
a
fact
or
cau
sin
g
t
he
po
s
sibil
it
y
of
f
ind
in
g
a
n
op
ti
mal
so
luti
on
t
ha
t
is
fast
an
d
a
ccur
at
e
is
getti
ng
small
er.
O
n
the
ot
her
ha
nd,
SCA
has
good
pro
w
ess
in
te
r
ms
of
global
sea
rch
capab
il
it
ie
s.
S
CA
ha
s
a
bala
nce
in
ex
plorat
ion
a
nd
e
xp
l
oitat
ion
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
P
ow Elec
& Dri S
ys
t
IS
S
N: 20
88
-
8
694
Impr
ovin
g neural
network
us
i
ng a sine
tree
-
s
eed
algorit
hm f
or
t
un
i
ng moto
r D
C
(
Wi
di
Aribo
w
o
)
1199
The
STS
A
al
gorith
m
i
nteg
ra
te
s
TS
A
with
SCA
t
o
imp
rove
t
he
abili
ty
to
fin
d
opti
mal
global
stoc
ha
sti
c
so
luti
ons i
n opt
imi
zat
ion
prob
le
ms.
T
he (
2
)
a
nd
()
in
the
TS
A
a
re r
e
place
d by
.
B
i
,
j
t
+
1
=
t
mpran
d
×
P
r
,
j
t
+
(
1
−
t
mpran
d
)
×
,
<
0
.
5
(14)
,
+
1
=
,
+
×
(
−
,
×
,
)
×
sin
(
×
a
cos
(
,
)
)
,
0
.
5
≤
<
0
.
5
(
15)
B
i
,
j
t
+
1
=
P
r
,
j
t
×
P
i
,
j
t
+
k
×
(
x
t
×
P
i
,
j
t
−
P
r
,
j
t
)
x
cos
(
π
x
a
rccos
(
x
t
)
)
,
0
.
5
≤
<
0
.
5
(
16)
Wh
e
re
is
a
ne
w
pa
rameter
ba
sed
on
the
or
i
gin
al
TS
A
.
T
he
s
up
e
rv
is
or
pa
rameter
f
or
c
on
t
ro
ll
in
g
t
he
search
tren
d
is
ST
.
High
e
r
ST
s
uppli
es
so
li
d
l
ocal
s
earch
a
nd
acce
le
rates
co
nv
e
r
gen
ce
,
a
nd
vic
e
ver
sa
lo
wer
values
le
ad
to
slo
w
c
onve
r
gen
ce
but
strong gl
ob
al
s
earch
.
2.2.
F
eed
-
fo
r
w
ard neural
net
w
or
k
Neural
netw
orks
a
re
mac
hin
e
le
arn
i
ng
dev
ic
es
ins
pirite
d
by
bio
l
og
ic
al
ne
ur
al
netw
orks
and
ca
pab
l
e
of
processin
g
with
mimi
c
a
s
the
huma
n
brai
n
[
21
]
.
Ar
ti
fici
al
ne
ur
al
netw
orks
(
A
NN
)
c
ould
c
on
st
ru
ct
model
li
near
an
d
nonl
inear
al
gorit
hms.
T
hey
a
re
great
ly
ans
we
re
d
as
pote
nt
tool
s
fo
r
t
he
op
ti
miza
ti
on
f
unct
ion
.
T
he
feed
-
f
orwa
rd
ne
ur
al
net
wor
k
(F
F
NN
)
is
t
o
appr
ox
imat
e
a
nd
c
on
t
ro
l
no
nl
inear
relat
ion
s
hip
s
betwee
n
i
nputs
and
outp
uts.
T
he
data
m
od
el
nee
ds
to
m
ove
in
onl
y
one
directi
on
(
f
orward)
f
rom
th
e
input
nodes,
via
the
hidden
laye
rs,
and eve
ntu
al
l
y t
o
the
outp
ut nod
e
.
F
FNN
has
a to
po
l
ogy w
hi
ch
co
ns
i
sts
of
process
unit
s, whic
h
are
re
pr
ese
nte
d
by
ne
uro
ns
.
F
igure
1
is
the
topo
lo
gy
of
F
F
NN.
Ne
uro
ns
are
the
m
os
t
i
mporta
nt
it
ems
of
t
he
FFNN
w
hich
a
re
gove
r
ned
by
in
put,
hidde
n
la
yer
s
a
nd
ou
t
pu
ts
.
The
in
pu
t
(
)
se
nds
a
sign
al
to
the
hidden
la
yer
th
rou
gh
a
wei
gh
te
d
net
work
(
W
ij
)
.
In
this
sect
io
n,
th
e
hi
dd
e
n
neuron
receive
s
a
w
ei
gh
t
pl
us
bias
in
pu
t.
The
n,
t
he neu
r
on
s
are
d
i
rected to
the
outp
ut
la
yer
(
4
)
.
F
FNN
can be
f
or
m
ul
at
ed
as
.
1
(
)
=
∑
(
t
)
+
1
=
1
(17)
2
(
)
=
(
1
(
)
)
=
1
1
+
1
(18)
3
(
)
=
∑
2
(
t
)
+
2
=
1
(19)
4
(
)
=
(
3
(
)
)
=
1
1
+
3
(20)
Fig
ure
1
.
FF
N
N
s
tr
uctu
re
2.3.
DC
mo
t
or
DC mo
t
or
is a
typ
e
of
mo
t
or
t
hat u
ses s
pee
d and
posit
ion
c
ontr
ol which is the arma
ture
c
ontr
ol o
f
t
he
DC
m
oto
r
[
22
]
.
A
rmatu
re
c
ontrol
by
keep
i
ng
the
sta
ti
c
fiel
d
cu
rr
e
nt
c
on
st
ant
is
ke
y
in
D
C
mo
to
r
set
up
[
23
]
.
The
e
quivale
nt
circuit i
s il
lust
rated i
n
Fi
gure
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
D
ri
S
ys
t,
V
ol
.
12
, N
o.
2
,
J
une
2021
:
119
6
–
120
4
1200
Fig
ure
2.
DC
mo
to
r
e
quivale
nt circ
uit [
24
]
Table
1.
DC m
otor
par
amet
e
r
s
[
25
]
No
tatio
n
Info
rm
atio
n
R
a
Armatu
re
resistan
c
e
L
a
Armatu
re
in
d
u
ctan
ce
e
a
Armatu
re
v
o
ltag
e
e
b
Back
electr
o
m
o
tiv
e f
o
rce
J
Moment o
f
intertia
con
stan
t
B
Dam
p
in
g
f
ri
ctio
n
r
atio
T
m
Moto
r
to
rqu
e
K
b
EM
F
co
n
stan
t
K
i
Moment co
n
stan
t
(
)
=
(
+
.
)
.
(
)
+
(
)
(21)
(
)
=
(
)
(22)
(
)
=
(
)
+
(
)
(23)
(
)
=
(24)
2.4.
The pr
oposed
STSA
-
N
N
m
ode
l
The
flo
w
cha
rt
of
t
he
ST
SA
-
NN
hybrid m
et
hod
is pr
ese
nte
d
in Figure
3
i
n
A
ppen
di
x
.
S
TSA
a
nd
N
N
work
in
dep
e
ndently
.
T
he
t
wo
proce
sses
furthe
r
ha
ve
i
nteracte
d
with
each
oth
e
r
to
form
t
he
S
T
SA
-
NN
method.
I
n
the
method
propo
sed
by
S
TS
A
-
NN,
data
f
r
om
the
ge
ner
at
or
is
ta
ken
as
in
put
an
d
ta
rget
f
or
NN
trai
ning.
The
da
ta
is
pr
ocesse
d
a
nd
gro
up
e
d.
N
N
is
set
us
in
g
se
ve
ral
par
a
mete
rs.
T
he
w
ei
gh
ti
ng
pa
ram
et
er
at
the
beg
i
nn
i
ng
will
be
use
d
a
rand
om
val
ue.
This
weig
ht
va
lue
will
be
inc
r
eased
us
i
ng
th
e
ST
SA
met
hod.
T
he
weig
hted res
ults fro
m ST
SA
will
b
e st
or
e
d
i
n
the
NN.
3.
RESU
LT
S
AND DI
SCUS
S
ION
In
the
first
ste
p
of
t
he
te
st
,
t
he
DC
m
otor
is
m
od
el
e
d
on
the
M
A
TLA
B/
Simuli
nk
(R2015a)
.
D
C
mo
to
r
modeli
ng
is
regulat
ed
us
in
g
th
e
P
ID
method.
P
ID
c
on
t
ro
ll
er
p
a
ra
mete
r
values
c
an
be
see
n
in
T
able
2.
The
nece
ssar
y
data
wer
e
ca
pt
ur
e
d
an
d
grouped
as
a
ref
e
re
nce
f
or
the
ST
SA
-
N
N
intel
li
gen
t
c
on
tr
ol
tr
ai
nin
g.
Schemati
c
of
DC
m
otor
c
on
t
ro
l
us
in
g
STS
A
-
NN
ca
n
be
s
een
in
Fi
gure
4
.
T
he
par
a
mete
rs
a
nd
val
ues
use
d
i
n
the
pro
posed
STSA
-
NN
method
ca
n
be
se
en
in
Ta
ble
3
.
The
c
onve
rg
e
nce
cu
r
ve
of
STSA
is
re
por
te
d
in
Figure
7
.
In
F
ig
ure
5
,
th
e
co
nver
ge
nce
value
of
STS
A
sto
ps
at
it
erati
on
25
.
T
his
re
su
lt
is
obta
ine
d
by
e
nteri
ng
the
par
a
mete
rs
co
ntaine
d
i
n
T
able
3
i
n
t
he
STS
A
-
NN
al
gorithm
.
T
he
nex
t
ste
p
,
it
is
to
i
ns
ta
ll
an
d
te
st
the
STSA
-
NN
co
nt
ro
ll
er
on
the
DC
M
ot
or
.
T
he
res
ult
of
a
cl
os
e
d
lo
op
s
ys
t
em
bet
ween
th
e
c
ontr
oller
a
nd
the
dc
mo
to
r
ca
n be s
een in Fi
gu
re
6. Ta
ble 4 is a
de
ta
il
o
f DC
M
ot
or
outp
ut w
a
ve
.
Table
2
. Para
m
et
er of
P
I
D
c
ontrolle
r
Para
m
eter
Valu
e
K
p
2
K
I
6
.5
K
D
0
.01
Table
3
. Para
m
et
er of
S
TS
A
-
NN
Para
m
eter
Valu
e
Hid
d
en
layer
4
Tr
ain
in
g
Leven
b
erg
-
m
a
rqu
ardt
Maximu
m
ite
ration
nu
m
b
er
50
Nu
m
b
er
o
f
p
o
p
u
latio
n
s
50
Lower bo
u
n
d
; up
p
er
b
o
u
n
d
-
1
.28
;1
.28
ST
0
.3
Up
p
er
lim
it
of see
d
's n
u
m
b
er
0
.1
Lower li
m
it
of see
d
's n
u
m
b
er
0
.25
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
P
ow Elec
& Dri S
ys
t
IS
S
N: 20
88
-
8
694
Impr
ovin
g neural
network
us
i
ng a sine
tree
-
s
eed
algorit
hm f
or
t
un
i
ng moto
r D
C
(
Wi
di
Aribo
w
o
)
1201
Fig
ure
4
.
D
C
mo
to
r
s
pee
d
c
ontr
ol w
it
h
STS
A
-
NN
co
ntr
oller
Figure
5
.
The
c
onve
rg
e
nce c
urve
of STS
A
Figure
6
.
S
pee
d
ste
p
re
spo
ns
e
of the
D
C
mo
t
or
In
Table
4
,
a
com
par
at
ive
a
nalysis
of
the
pro
po
se
d
STS
A
-
NN
ap
proac
h
c
ompare
d
to
PID
is
s
how
n
.
T
o
determi
ne
the
eff
ect
ive
ness
a
nd
validat
io
n
of
the prop
os
ed
method,
tw
o
c
r
it
eria
wer
e u
se
d
, n
amel
y
In
te
gr
al
o
f
ti
me
mu
lt
ipli
ed
by
s
qu
a
re
d
e
rror
(I
T
SE)
a
nd
I
nteg
ral
of
ti
me
mu
lt
ipli
ed
by
a
bsolute
er
r
or
(I
T
AE
).
ITSE
has
a
su
pple
me
ntar
y
ti
me
mu
lt
ipli
er
of
th
e
er
ror
functi
on
wh
ic
h
po
i
nts
on
t
he
sp
a
n
of
t
he
error
durati
on.
This
measu
reme
nt i
s popular
in
t
he
sy
ste
m
takes
qu
ic
k
tu
nnin
g
t
ime. T
he
I
TSE
ind
e
x f
or
m
ula
is give
n
as
.
=
∫
.
2
(
)
.
∞
0
(25)
ITAE
is
integ
r
at
ing
t
he
a
bsol
ute
e
rror
m
ulti
plied by
ti
me
a
fter
ti
me.
M
ini
mizi
ng
the
inte
gr
al
of
ti
me
-
w
ei
gh
te
d
abs
olu
te
er
r
or
(
ITAE) is
as
usual
pointe
d
t
o
as a
good ac
hieveme
nt in
de
x
i
n plott
ing c
on
t
ro
ll
ers
.
=
∫
.
(
)
.
∞
0
(26)
The
c
omparis
on
of
t
he
ITAE
and
IT
SE
ca
n
be
see
n
i
n
Ta
bl
e
4.
The
IT
AE
value
of
t
he
P
ID
has
val
ue
0.594
4.
M
ea
nwhile
,
th
e
lowest
IT
AE
value
is
own
e
d
by
t
he
ASO
-
NN
meth
od.
It
is
0.335
7.
T
he
lowest
ITSE
va
lue
is
owne
d
by
t
he
STSA
-
NN
met
hod
of
0.1
423.
O
n
the
oth
er
hand,
the
high
est
val
ue
of
IT
SE
is
ow
ned
by
t
he
M
P
A
-
N
N
met
hod o
f 0.
2057
Table
4.
C
omp
ariso
n of t
he
tr
ansient
res
ponse
an
al
ysi
s r
es
ul
ts for
dif
fer
e
nt
con
t
ro
ll
ers
.
Co
n
troller
Ov
ersh
o
o
t
Settlin
g
T
im
e (
s
)
Ris
e T
i
m
e (
s
)
IT
SE
IT
A
E
PID
1
.00
0
3
0
.13
7
5
0
.0
81
0
.20
0
7
0
.59
4
4
FFNN
No
ov
ersh
o
o
t
0
.13
7
5
0
.08
0
.20
2
5
0
.71
5
5
MPA
-
NN
No
ov
ersh
o
o
t
0
.15
0
.08
5
0
.20
5
7
0
.73
6
4
ASO
-
NN
1
.04
2
0.1
3
0.0
4
0
.15
1
5
0
.33
5
7
STSA
-
NN
No
o
v
ersh
o
o
t
0.
11
0
.0
55
0
.14
2
3
0
.40
9
7
4.
CONCL
US
I
O
N
M
a
nag
i
ng
co
nt
ro
l
on
a
DC
m
otor
is
a
very
interest
ing
res
earch
area
.
D
C
mo
t
or
c
ontr
ol
that
has
a
set
po
i
nt
value
mu
st
pa
y
at
te
nt
ion
t
o
t
he
ef
fici
ency
of
the
con
t
ro
l.
T
he
sine
t
ree
-
see
d
al
gorithm
a
nd
ne
ur
al
netw
ork
hybri
d
met
hod
pro
pose
d
in
this
study
is
use
d
to
con
t
ro
l
DC
m
otors.
From
th
e
resea
rch
res
ul
ts,
the
pro
po
se
d
met
hod
si
ne
tree
s
eed
al
gorith
m
-
neural
netw
or
k
(STSA
-
NN)
ha
s
a
bette
r
performa
nce
on
t
he
set
tl
ing
ti
me
a
nd
I
TAE
par
a
mete
rs
t
han
th
e
pe
rfo
rma
nce
of
P
ID.
T
he
IT
AE
val
ue
of
th
e
pro
pose
d
me
thod
i
s
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
D
ri
S
ys
t,
V
ol
.
12
, N
o.
2
,
J
une
2021
:
119
6
–
120
4
1202
0.409
7
.
M
ea
nwhile
,
t
he
ITSE
val
ue
of
the
pro
po
se
d
meth
od
ha
s
bette
r
pe
rformance
tha
n
the
P
ID
val
ue
.
It
is
0.142
3.
T
he
l
owest
IT
AE
val
ue
is
owne
d
by
the
A
SO
-
NN
method
of
0.3
357.
H
ow
e
ve
r,
the
ASO
-
N
N
m
et
hod
has
a
n
ove
rshoo
t
of
1.0
423.
To
te
st
the
pe
rformance
of
the
ST
SA
-
N
N
method,
it
is
ne
cessar
y
to
ca
r
ry
out
furthe
r
re
sear
c
h wit
h
m
ore c
omplex
pr
ob
le
m
s.
APPE
ND
I
X
Fig
ure
3
.
The
pro
po
se
d STS
A
-
NN fl
ow
c
ha
rt
REFERE
NCE
S
[1]
A.
Lot
fy
,
M
.
Ka
veh,
M
.
R.
Mos
avi
and
A.
R.
R
ahm
a
ti
,
"
An
enh
anc
ed
fuz
zy
con
trol
ler
base
d
on
im
prove
d
geneti
c
al
gorit
h
m
for
sp
ee
d
cont
ro
l
of
DC
mot
ors,
"
An
alog
Int
egr
Circ
Sig
Proce
ss
,
vo
l.
105
,
pp.
141
-
155
,
2020
,
DOI
:
10.
1007/s10470
-
020
-
01599
-
9
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
P
ow Elec
& Dri S
ys
t
IS
S
N: 20
88
-
8
694
Impr
ovin
g neural
network
us
i
ng a sine
tree
-
s
eed
algorit
hm f
or
t
un
i
ng moto
r D
C
(
Wi
di
Aribo
w
o
)
1203
[2]
Bapa
yya
Naidu
Kommul
a
,
and
Venka
ta
Reddy
Kota,
"
Dire
c
t
in
stant
an
eous
torq
ue
cont
rol
of
Br
ushless
DC
mo
t
or
using
fir
efl
y
Al
gorit
hm
bas
ed
f
rac
t
iona
l
ord
er
PID
cont
rol
le
r
,
"
Journal
o
f
Kin
g
Saud
Univ
ersity
-
Engi
ne
ering
Sci
en
ce
s
,
v
ol.
32
,
no
.
2,
p
p
.
133
-
140,
2020
.
[3]
Obed
A
.
A
.
,
Sal
eh
A
.
L
.,
and
K
adhi
m
A
.
K
.
,
"
Speed
per
for
ma
n
ce
ev
al
ua
ti
on
of
BLDC
mot
or
b
ase
d
on
dynami
c
wave
let
neur
al
net
work
and
PS
O
al
gorit
hm
,
"
Inte
rnationa
l
J
ournal
of
Pow
e
r
El
e
ct
ronics
a
nd
Dr
iv
e
S
ystem
(IJ
PE
DS)
,
vol
.
1
0,
no
.
4
,
pp
.
174
2
-
17
50,
2019
,
D
OI
:
10.
11591/
ij
p
eds.
v10.
i4
.
1742
-
1750.
[4]
W.
Aribowo
,
S.
Mus
li
m
and
I
.
Basuki,
"
Gene
r
al
i
ze
d
r
egr
essio
n
neur
a
l
n
et
wor
k
for
long
-
te
r
m
elec
tr
icity
loa
d
fore
ca
st
ing
,
"
20
20
Int
ernati
ona
l
Con
fe
renc
e
o
n
Smar
t
Techn
ology
and
App
li
cations
(ICoS
TA)
,
Sur
aba
ya
,
Indone
sia, pp. 1
-
5,
2020
,
DOI
:
10
.
1109/ICoSTA48221.
2020.
1570
611361.
[5]
Yasar,
A.
,
Sar
itas,
I
.
,
Sahma
n
,
M.A.,
Dundar
,
A.O,
"
Cla
ss
ifica
ti
on
of
leaf
type
using
ar
ti
fi
cial
neur
al
net
works
,
"
Int.
J. I
nt
el
l
.
S
yst
.
App
l. E
ng
.
v
ol
.
3
,
no
.
4
,
pp
.
136
-
139,
2015
,
DOI
:
10.
18201
/i
j
isae.49279
.
[6]
Sulistyo.
S.B,
W
oo.
W.L,
and
Dl
ay.
S.S,
"
Regula
riz
ed
n
eur
al
n
etw
orks
fusion
an
d
gen
et
i
c
a
lgori
t
hm
b
ase
d
on
-
fi
eld
nit
roge
n
st
at
us
esti
mation
of
whea
t
p
la
n
ts,
"
IE
EE
Tr
ans.
Indu
str.
Inf
.
vo
l.
13
,
no.
1
,
pp.
103
-
114,
2016
,
DOI
:
10.
1109/T
II
.
201
6.
2628439
.
[7]
Gu.
K,
Zhou.
Y,
Sun.
H
,
Zh
ao.
L,
and
Li
u
.
S,
"
Predic
ti
on
of
a
ir
qua
li
ty
in
Sh
enz
hen
base
d
o
n
neur
a
l
n
et
w
or
k
al
gorit
h
m,
"
Neur
al
Comput.
App
l
.
,
vol.
32,
pp.
18
79
-
1892
,
2020
,
DOI
:
10.
1007/s0
0521
-
019
-
04492
-
3
.
[8]
B.
Heki
moğl
u,
S.
Eki
n
ci
,
and
S.
Kaya
,
"
Opti
ma
l
PID
con
troller
d
esign
of
D
C
-
DC
buck
con
ver
te
r
using
wh
al
e
opti
mization
a
lgori
thm,
"
in
Proc.
I
EEE
IDAP
,
Ma
l
at
ya
,
Turk
ey,
Sep.
2018,
pp.
1
-
6
,
DOI
:
10.
1109/IDAP
.
2018.
8620833
.
[9]
B.
Nay
ak
and
S.
Sahu,
"P
aramet
er
est
imati
on
of
DC
mot
or
throu
gh
whal
e
optimi
za
t
ion
al
gor
it
hm
,"
In
te
rnationa
l
Journal
of
Po
wer
Elec
troni
c
s
and
Dr
iv
e
Syste
m
(IJ
PE
D
S)
,
v
ol
.
10
,
n
o.
1,
pp.
83
-
92,
2019
,
DOI
:
10.
11591/ijpeds.
v10.
i1.
pp83
-
92
.
[10]
G.
Na
lc
a
ci,
D.
Yildi
rim
and
M.
Ermis,
"S
el
e
ctive
Harm
on
ic
E
li
mi
n
at
ion
for
L
ight
-
Rail
Tr
ansporta
ti
on
Motor
Drive
s
using
H
arr
is
Hawks
Al
gorit
hm
,
"
2020
IEE
E
Inte
rnat
io
nal
Confe
r
enc
e
on
En
vi
r
onme
nt
and
El
e
ct
ric
al
Engi
ne
ering
and
2020
IE
EE
Indu
strial
and
Comm
erc
ial
Pow
er
S
yste
ms
Europe
(
E
EE
IC
/
I&C
PS
E
urope)
,
Madrid
,
Spain,
2020
,
pp
.
1
-
6,
DOI
:
10
.
11
09/E
EEIC/ICPS
Europe
49358.
20
20.
9160694.
[11]
D.
Puangdownreong,
S
.
Hlung
nam
ti
p
,
C
.
Thamma
r
at
and
A
.
Naw
ika
vatan,
"
Applic
a
ti
on
of
f
lower
pol
li
na
ti
o
n
al
gorit
h
m
to
par
am
e
te
r
ide
nt
ifi
c
at
ion
of
DC
mo
tor
model,
"
201
7
Inte
rnationa
l
El
e
ct
rica
l
Eng
in
ee
ring
Congress
(iE
ECON)
,
Patta
ya,
2017
,
pp
.
1
-
4
,
DOI
:
10
.
1109/I
EE
CON
.
2017.
8
075889.
[12]
B.
N.
Reddy
K
ota
,
"
Dire
ct
inst
ant
a
n
eous
torqu
e
cont
rol
of
b
ru
shless
DC
mot
o
r
using
fire
f
ly
a
lg
orit
hm
b
ase
d
fra
ctional
ord
er
PID
cont
roll
er
,
"
Journal
of
Kin
g
Saud
Unive
rs
i
ty
-
Eng
ine
ering
Sci
en
ce
s
,
vo
l.
3
2,
no.
2
,
pp.
13
3
-
140
,
Februa
ry
2
020
,
DOI
:
10
.
10
16/j
.
jksu
es.
2018.
04.
007
.
[13]
E.
C.
Şimşek
,
A
.
Kös
e,
M.
Şahi
n
and
E
.
Irma
k
,
"O
pti
mi
z
at
ion
o
f
PID
par
am
e
te
r
s
using
an
t
col
o
ny
al
go
rit
h
m
for
positi
on
cont
ro
l
of
DC
m
o
tor,"
2019
8
th
Int
ernati
onal
Conf
ere
nce
on
R
en
ewabl
e
En
ergy
R
ese
arch
and
Appl
ic
a
ti
ons (IC
RE
R
A)
,
Brasov,
Roma
nia,
2019
,
pp.
1047
-
1051
,
DOI
:
10.
1109/I
CRERA47325.2019.
8997108.
[14]
B.
A.
Kouass
i,
Y.
Zha
ng
,
S.
Ou
at
t
ara
and
M.
J.
Mbyamm
Kiki
,
"
PID
tuni
ng
of
choppe
r
f
ed
spe
ed
cont
ro
l
of
D
C
mot
or
base
d
o
n
ant
col
ony
o
pti
mization
al
g
orit
hm
,
"
2019
IEE
E
3rd
In
te
r
nati
onal
El
e
ct
ri
cal
and
Ene
rg
y
Confe
renc
e
(CI
E
EC)
,
B
ei
j
ing, Ch
ina
,
2019,
pp.
40
7
-
412,
DOI
:
10
.
1109/CIE
EC471
46.
2019.
CIE
EC
-
2019179.
[15]
C.
Fu,
Q.
-
G.
W
ang,
J.
Yu
and
C.
L
in,
"N
eur
a
l
net
work
-
base
d
fi
nit
e
-
t
ime
com
m
and
fi
ltering
con
trol
for
sw
it
ch
ed
nonli
ne
ar
sys
te
m
s
with
bac
k
la
sh
-
l
ike
hyster
esis
,
"
in
IEE
E
Tr
ansacti
ons
on
Neural
Net
works
and
Le
arning
Syste
ms
,
2020,
pp
.
1
-
6
,
D
OI
:
10.
1109/
TNNLS.2020.
3009
871.
[16]
C.
Fu,
L
.
Zh
ao,
J.
Yu,
H.
Yu,
a
nd
C.
L
in,
"N
eu
ral
ne
twork
-
base
d
com
m
and
fi
lt
e
red
cont
ro
l
for
i
n
duct
ion
mot
ors
with
inpu
t
satu
rat
ion
,
"
I
ET
Co
ntrol
Theory
a
nd
App
li
ca
ti
ons
,
vo
l.
11,
no
.
15,
p
p
.
2636
-
2
642,
2017
,
DOI
:
10.
1049/iet
-
c
ta.2
017.
0059
IE
T
[17]
S.
Gajbhi
ye
,
M.
A.
Ghorbani
,
S
.
S.
Band,
V.K
.
Khos
rows
hahi
and
C.
Meshram
,
"
Rive
r
flow
pr
edicti
on
using
hybr
i
d
PS
OGSA
al
gori
thm
b
ase
d
on
f
ee
d
-
forward
neu
ral
net
work
,
"
So
ft
Comput
ing
,
v
ol.
23,
no.
3
,
p
p.
10429
-
10438
,
2019
,
DOI
:
10
.
1
007/s00500
-
018
-
3598
-
7
.
[18]
X.J.
Luo
,
Luku
mon
O.O
.
,
Anuo
luwapo
O.A
.
,
Ol
ugbenga
O.A
.
,
Delga
d
o
J.M
.
D,
and
Hake
em
A.
O
.
,
et
al.
,
"
Gene
t
ic
al
gorit
h
m
-
de
te
r
mi
ned
de
ep
fe
ed
forward
n
eur
al
net
work
arc
h
it
e
c
ture
for
pre
d
ic
t
i
ng
e
lectr
i
ci
ty
co
nsumpti
on
in
re
al
buil
dings,
"
Ene
r
gy
and
AI
,
v
ol
.
2
,
2020
,
DOI
:
10
.
1016/j
.
egya
i
.
202
0.
100015
.
[19]
Garc
ía
-
Ród
ena
s,
R.
,
Li
nar
es,
L
.
J.
and
Lóp
ez
-
G
óme
z
.
J.A,
"
Me
m
et
i
c
a
lgori
th
m
s
for
tra
ini
ng
fe
edf
orward
neur
a
l
net
works
:
an
ap
proa
ch
b
ase
d
on
gra
vitat
iona
l
se
arc
h
al
gori
thm,
"
Neural
Comput
&
App
li
c
,
vo
l.
33
,
no.
1
,
2020
,
DOI
:
10.
1007/s0
0521
-
020
-
05131
-
y
.
[20]
Jianhua
Ji
ang,
Meirong
Xu,
Xi
anqi
u
Meng
,
K
e
qin
L
i,
"
STS
A:
A
sine
tre
e
-
s
ee
d
al
gor
it
hm
for
c
ompl
ex
continuo
u
s
opti
mization
pr
oble
ms
,
”
Phy
si
ca
A:
Sta
ti
sti
c
al
Me
chanics
and
it
s
Appl
i
c
ati
ons
,
vo
l
.
53
7,
2020
.
DOI
:
10.
1016/j.physa.2019.122802
.
[21]
W.
Aribowo,
S.
Mus
li
m,
muno
t
o,
B.
Supria
nto,
U.
T.
Kar
ti
ni
a
nd
I.
G.
P.
As
to
Budit
j
ah
ja
n
to,
"
Tuni
ng
of
powe
r
sys
te
m
stabi
l
ize
r
using
ca
sca
d
e
forward
bac
kpr
opaga
t
io
n,
"
202
0
Thir
d
Inte
rnational
Confe
ren
c
e
on
Voc
a
ti
onal
Educ
ati
on
an
d
Elec
tri
cal
Engi
n
ee
ring
(ICVEE)
,
Suraba
ya,
Ind
onesia
,
2020,
pp.
1
-
5,
DOI
:
10.
1109/ICVEE50212.2020.
924
3204.
[22]
Ram
ír
ez
-
Cárd
en
as,
O
.
-
D.
,
and
Truj
illo
-
Rom
ero
,
F,
"
Sensorl
ess
spee
d
tracki
ng
of
a
brushl
ess
DC
mot
or
using
a
neur
al ne
twork
,
"
Math. Comput.
Appl
,
vol
.
25
,
no
.
57
,
2020
,
DOI
:
10.
3390/mca250
30057
.
[23]
Badri
yah
Ahm
e
d
Obaid,
Am
ee
r
La
t
ee
f
Sa
le
h
,
a
nd
Abbas
Kare
e
m
Kadhim,
"
Re
solving
of
optim
al
fra
ct
ion
al
PI
D
cont
roller
for
D
C
mot
or
driv
e
b
ase
d
on
anti
-
win
dup
by
inva
sive
wee
d
optimization
te
chn
ique
,
"
I
ndonesian
Journ
al
of
Elec
tric
al
Engi
ne
ering
an
d
Computer
S
ci
en
ce
(
I
JEECS
)
,
v
ol
.
15
,
no
.
1,
pp.
95
-
1
03
,
2019
,
DOI
:
10.
11591/ijeecs.
v15.
i1.
pp95
-
103
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8
694
In
t J
P
ow
Ele
c
&
D
ri
S
ys
t,
V
ol
.
12
, N
o.
2
,
J
une
2021
:
119
6
–
120
4
1204
[24]
S.
J.
Hamm
ood
i,
K.S.
Flayyi
h
,
a
nd
A.R.
Ham
ad,
"
Design
and
implementation
sp
ee
d
con
trol
sys
t
em
of
DC
Moto
r
base
d
on
PID
cont
rol
and
Ma
tl
ab
Simul
ink,
"
Int
ernati
onal
J
ournal
of
P
ow
e
r
Elec
troni
cs
a
nd
Dr
iv
e
Syst
e
m
(IJ
PE
DS)
,
vol
.
1
1,
no
.
1
,
pp
.
127
-
134
,
Mar
ch
202
0
,
DOI
:
10
.
1159
1/i
jpe
ds
.
v11.
i1
.
p
p127
-
134
.
[25]
Amiri
,
M.S.
,
Ibr
ahi
m
,
M.F
.
,
Ra
ml
i,
R
,
"
Opti
mal
pa
ramet
er
estimat
ion
for
a
D
C
mot
or
using
g
ene
t
ic
al
gor
it
hm
,
"
Inte
rnational
Jo
urnal
of
Pow
er
El
e
ct
ronics
and
Dr
iv
e
S
yste
m
(IJ
PE
DS),
vol
.
11,
no.
2
,
pp.
1047
-
1054,
2020
,
DOI
:
10.
11591/ijpeds.
v11.
i2.
pp1047
-
1
054
.
BIOGR
AP
HI
ES OF
A
UTH
ORS
Widi
Aribowo
i
s
a
lectur
er
in
t
he
Dep
art
m
ent
of
Elec
tri
c
al
En
gine
er
ing,
Univ
ersi
ta
s
N
ege
ri
Suraba
ya,
Indo
nesia
.
He
is
B
.
Sc
in
Pow
er
Engi
ne
eri
ng/Sep
uluh
Nopemb
er
Instit
u
te
of
Te
chno
logy
(IT
S)
-
Suraba
ya
in
2005.
He
is
M.
Eng
in
Pow
er
E
ngine
er
ing/
Sepu
luh
Nopembe
r
Instit
ute
of
T
echnology
(IT
S)
-
S
ura
baya
in
2009
.
He
is
mainly
r
ese
arc
h
in
th
e
p
ower
sys
te
m
and
con
trol.
Bamban
g
Su
pr
ianto
is
a
l
ectur
er
in
th
e
Dep
artme
nt
of
El
e
ct
r
ical
Engi
ne
eri
ng
,
Univer
sita
s
Nege
ri
Sur
aba
ya
,
Indon
esia
.
He
com
pl
et
ed
Ba
ch
el
or
of
E
lectr
oni
c
Eng
ineeri
ng
E
duca
t
ion
in
Univer
sita
s
Neg
eri
Suraba
y
a
-
Suraba
ya
in
1986.
He
holds
Master
Engi
n
ee
ring
in
Sepuluh
Nopembe
r
I
nsti
t
ute
of
T
ec
hnolo
gy
(IT
S)
-
Surab
a
ya
in
2001.
He
was
com
p
le
t
ed
Doctor
of
El
e
ct
ri
ca
l
Engi
n
ee
ring
in
Sepulu
h
Nopembe
r
Inst
it
ute
of
Technol
ogy
(IT
S)
-
Suraba
ya
in
2012
.
His re
sea
r
ch
in
terests i
nc
lude po
wer
sys
te
m, c
on
t
rol
and
e
le
c
tronic.
Jok
o
is
a
le
c
tur
er
in
th
e
Depa
r
t
me
nt
of
El
e
ct
r
ical
Eng
ine
er
ing,
Univer
sita
s
Neg
eri
Suraba
y
a,
Indone
sia.
He
c
ompl
eted
B
ac
he
l
or
of
Elec
troni
c
Engi
ne
eri
ng
Edu
ca
t
ion
in
Univ
er
sita
s
Nege
r
i
Suraba
ya
-
Surab
aya
in
1989
.
He
holds
Master
E
ngine
er
ing
in
Se
puluh
Nopember
I
nstit
ut
e
o
f
Te
chno
logy
(IT
S)
-
Suraba
ya
in
2
00
4
.
He
was
co
mpl
eted
Doctor
i
n
Univer
sita
s
Ne
ger
i
Mal
ang
-
Mala
ng
in
2017
.
His re
se
arc
h
intere
sts in
cl
ude
po
wer
sys
te
m
and mot
or
.
Evaluation Warning : The document was created with Spire.PDF for Python.