Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
10
,
No.
3
,
June
201
8
,
pp
.
1080~
1089
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v1
0.i
3.
pp
1080
-
1089
1080
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Optimi
zation
of
PV Syste
ms Usin
g Data Mi
nin
g
and Reg
ression
Learner
MPPT T
ec
hn
iqu
es
Ad
ed
ayo M
. F
aray
ola
,
Ali
N
Hasan
,
Ah
me
d A
li
Depa
rtment
o
f
E
le
c
tri
c
al E
ngin
eering
T
ec
hnolog
y
,
Univer
si
t
y
of
Johanne
sburg,
South Afri
c
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
N
ov
1
9
, 201
7
Re
vised
J
an
21
, 201
8
Accepte
d
Ma
r
11
, 201
8
Supervised
m
achine
l
ea
rn
ing
t
e
chni
ques
such
a
s
art
ifici
al
n
eur
al
ne
twork
(AN
N)
and
ANFIS
are
powerful
tool
s
used
to
tra
c
k
the
m
axim
um
po
wer
point
(MP
PT)
in
photovoltaic
s
y
stems
.
How
eve
r,
th
ese
offl
ine
MP
PT
te
chn
ique
s
stil
l
req
uire
l
arg
e
an
d
ac
cur
a
te
tr
ai
n
i
ng
dat
a
sets
for
succ
essful
tra
ck
ing
.
Th
is
pa
per
pre
sents
an
i
nnovat
iv
e
use
of
rat
ional
quadr
atic
gaus
sian
proc
ess
reg
ression
(RQGPR)
t
ec
hniqu
e
to
gene
rate
the
la
rg
e
and
v
e
r
y
ac
cur
ate
tra
in
ing
dat
a
req
uir
ed
for
MPPT
ta
sk.
To
conf
irm
the
eff
ec
t
ive
ness
of
the
RQG
PR
t
ec
hniqu
e,
the
co
m
bina
ti
on
of
A
NN
and
RQG
P
R
as
AN
N
-
RQG
PR
te
chn
i
que
result
s
were
compare
d
with
the
conve
nt
i
onal
AN
N
te
chn
ique
r
esults
,
and
that
of
combined
AN
N
and
li
ne
ar
sup
port
ve
ct
or
m
ac
hine
reg
r
ession
as
ANN
-
LSV
M
te
chni
que
res
ult
s
under
diffe
r
ent
wea
th
er
condi
ti
ons
.
Result
s
show
tha
t
AN
N
-
RQGPR
te
chni
que
produc
ed
t
he
over
a
l
l
best
resul
t and
w
it
h
an
improved
per
form
anc
e
.
Ke
yw
or
d
s
:
ANN
Data m
ining
te
chn
i
qu
e
MPPT
Photo
vo
lt
ai
c s
yst
e
m
Suppor
t
v
ect
or m
achine
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ali N. Has
an
,
Dep
a
rt
m
ent o
f El
ect
rical
En
gi
neer
i
ng
Tec
hnology
,
Un
i
ver
sit
y
of J
oh
a
nnes
burg
,
P.O. Bo
x 5
24, Auc
klan
d Par
k 2
006, S
ou
t
h A
fr
ic
a
.
E
m
a
il
: alin@
uj
.ac.za
1.
INTROD
U
CTION
Photo
vo
lt
ai
c
(
PV
)
s
olar
e
nergy
is
a
cat
ego
r
y
of
su
sta
ina
bl
e
energy
that
ge
ner
at
es
it
s
ene
rg
y
nat
ur
al
ly
from
su
nligh
t.
This
form
of
energy
is
the
do
m
inant
ty
pe
of
re
new
a
bl
e
energy
so
ur
ce
as
PV
ene
rg
y
is
consi
der
e
d
to
be
le
ss
poll
utive,
ine
xhausti
ble,
noise
f
ree,
and
read
il
y
avail
able
com
par
ed
to
t
he
f
ossi
l
fu
el
e
nergy
[1
]
.
T
he
PV
cel
ls
conver
t
li
gh
t
ene
r
gy
into
el
ect
rical
or
heat
ene
r
gy.
H
ow
e
ve
r,
these
P
V
cel
ls
us
ua
ll
y
pro
du
ce
lo
w
a
nd
a
bnor
m
al
powe
r
w
hen
us
ed
with
ou
t
a
work
i
ng
m
axim
u
m
po
wer
point
trac
ki
ng
(
MPPT)
con
t
ro
ll
er
[
2].
In
sta
nd
-
a
l
on
e
PV
syst
e
m
s,
MPPT
c
on
t
ro
ll
ers
a
re
us
e
d
to
en
ha
nce
the
fl
ow
of
powe
r
from
the
connecte
d
P
V
cel
ls
to
the
con
necte
d
loa
d.
I
n
batte
ry
-
c
onne
ct
ed
PV
syst
em
s,
MPPT
te
chn
i
qu
e
s
are
use
d
to
protect
batte
rie
s
f
ro
m
ov
e
r
-
c
ha
rg
i
ng
an
d
dee
p
discha
rg
e
of
powe
r
f
r
om
the
cel
ls
[
3
]
.
Als
o,
MP
PT
te
c
hniqu
e
s
can
i
ncr
ease
th
e
eff
ic
ie
ncy
of
PV
cel
ls
du
ring
col
d
te
m
per
at
ur
e
a
nd
cl
oudy
days
w
he
n
the
s
un
ir
radi
ance
is
low
[
4].
MPPT
te
chn
i
qu
es
are
cl
assif
ie
d
into
th
ree
cat
egories:
o
n
li
ne,
offli
ne,
and
hybri
d
MPPT
te
chn
iq
ues
[5]
-
[
10
]
.
Re
cent
researc
h
has
a
dopte
d
the
us
e
of
gl
ob
al
op
ti
m
iz
ation
te
ch
niques
su
ch
a
s
ge
net
ic
al
go
rithm
(GA),
par
ti
cl
e
swar
m
op
ti
m
izati
on
(P
S
O),
s
i
m
ulate
d
an
ne
al
ing
(SA),
an
t
bee
c
olony
(
ABC)
an
d
im
pr
ove
d
conve
ntion
al
MPPT
te
ch
ni
ques
s
uc
h
as
m
od
ifie
d
Pe
rturb
&Obser
ve
an
d
m
od
ifie
d
inc
re
m
ental
cond
uc
ta
nce
to
i
m
pr
ove
the
tra
ckin
g
eff
ic
ie
nc
y
of
MPPT
co
ntr
ollers
in
PV
syst
e
m
s
[5
]
,
[1
1
]
.
N
eve
rthel
ess,
few
wor
k
is
done
us
in
g
data
m
ining
te
ch
niques
to
track
the
m
axim
u
m
p
ow
er
po
i
nt
in
PV
sy
stem
s.
Data
m
i
ning
is
a
com
pu
ti
ng
and
sta
ti
sti
c
al
too
l
us
e
d
to
di
scov
e
r
patte
r
ns,
rem
ov
e
no
is
e,
extract
inf
orm
ation
f
r
om
la
rg
e
data
set
s
,
an
d
conve
rsion
of
filt
ered
data
set
s
into
a
lo
gical
structu
re
fo
r
furthe
r
use
.
Data
m
ini
ng
c
om
bin
es
m
achine
le
arn
in
g,
sta
ti
sti
cs,
an
d
data
base
syst
e
m
s
into
a
n
al
l
-
in
-
on
e
te
ch
nique
.
Tasks
s
olv
e
d
us
in
g
data
m
ining
te
chn
iq
ues
are
broa
dly
di
vid
ed
into
six
cat
egories:
cl
ust
er
analy
sis,
ano
m
al
y
detect
ion
,
cl
assi
ficat
ion,
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
Op
ti
miz
atio
n o
f PV Sy
ste
ms U
sing D
ata Mi
ni
ng an
d
Re
gr
es
sion Le
ar
ne
r…
(
Adeday
o
M
. F
arayol
a
)
1081
su
m
m
arization,
regressi
on,
and
ass
ocia
ti
on
ru
le
le
arn
i
ng
(d
e
pe
nd
e
ncy
m
od
el
li
ng
)
[
12
]
.
Cl
us
te
r
ana
ly
sis
is
us
e
d
to
le
ar
n
the
sim
il
arit
ies
that
exist
in
data
set
s.
A
no
m
al
y
detection
detect
s
th
e
error
i
n
dat
a
set
s.
Cl
assifi
cat
ion
si
m
plifie
s
the
known
str
uctur
e
in
new
da
ta
set
s.
Su
m
m
arizat
ion
off
ers
an
ext
ra
com
pact
represe
ntati
on
of
the
data
set
s
w
hile
re
gr
e
ssion
le
ar
ning
est
i
m
at
es
the
rela
ti
on
s
hip
with
da
ta
set
s
an
d
at
tem
pts
to
find
a
f
un
ct
ion
that
m
od
el
s
the
data
set
s
with
le
ast
err
or.
Re
gr
e
ssio
n
le
arn
i
ng
us
es
predict
ive
analy
sis
to
ob
ta
in
m
or
e
accurate
res
ul
ts
fr
om
a
decisi
on
s
upport
syst
e
m
.
Dep
end
e
ncy
m
od
el
li
ng
exam
ines
the
relat
ion
s
hip
s
wi
th v
a
riables
[13
].
The
c
ontrib
ution
of
t
his
pa
pe
r
is
to
intr
oduc
e
the
us
e
of
a
par
ti
cula
r
cl
ass
of
data
m
ining
te
ch
nique
s
known
as
regr
ession
le
ar
nin
g
al
gor
it
hm
fo
r
op
ti
m
iz
a
ti
on
an
d
im
pr
ove
m
ent
of
MP
PT
co
ntr
ollers
in
PV
syst
e
m
s.
The
r
egr
es
sio
n
le
arni
ng
alg
or
it
hm
s co
nsi
der
e
d
are
li
near
sup
port v
ect
or m
achine (
LSV
M
)
re
gr
e
ssio
n
te
chn
iq
ue
a
nd
rati
onal
qua
dr
at
ic
gau
s
sia
n
process
re
gressi
on
(RQ
G
PR)
te
ch
nique
.
T
he
se
re
gre
ssion
al
gorithm
s
us
e
few
real
-
tim
e
sam
ples
to
gen
erate
the
ne
w
data
set
s
need
e
d
to
trai
n
the
ANN
f
or
MP
P
T
ta
sk
.
Seco
nd
c
on
tri
bu
ti
on
is
a
wo
r
k
done
to
e
valuate
the
fe
asi
bili
ty
of
the
two
pro
po
se
d
regressi
on
l
earn
i
ng
al
gorithm
s
(LSV
M
a
nd
R
Q
GP
R)
t
hat
we
re
com
bin
ed
s
epar
at
el
y
with
ANN
as
A
N
N
-
L
SV
M
a
nd
ANN
-
RQGP
R
te
ch
ni
qu
e
res
pecti
ve
ly
for
the
en
han
cem
ent
of
MPPT
te
chn
i
ques
in
P
V
syst
e
m
s
un
der
dif
fer
e
nt
weathe
r
c
onditi
on
s
.
The
syn
opsis
of
this
p
a
pe
r
is
pr
e
par
e
d
as
fo
ll
ows,
sect
io
n
2
will
pr
ese
nt
a
su
m
m
ary
of
the
use
d
MPPT
te
ch
niques.
In
sect
io
n
3,
a
re
port
of
the
ex
pe
rim
e
nt
set
up
a
nd
m
et
ho
d
is
pro
vid
e
d.
Sect
ion
4
will
pr
ese
nt the
r
es
ults, a
nd secti
on
5 wil
l i
nclu
de
the c
on
cl
us
io
ns
.
2.
MPPT T
E
CHNIQ
UES
The
te
c
hn
i
qu
e
s
u
se
d
i
n
this
stud
y a
re
br
ie
fly
discusse
d
:
2.1
.
Li
ne
ar
-
S
upp
ort
Vec
to
r
Machine
(LS
VM)
Suppor
t
vect
or
m
achine
(SV
M)
is
a
po
pu
la
r
m
achine
le
ar
ning
te
c
hn
i
qu
e
us
e
d
f
or
cl
ass
ific
at
ion
a
nd
regressio
n
a
nal
ysi
s.
SV
M
re
gressi
on tech
nique
was fir
st i
nt
rod
uced by
Vladim
ir V
a
pn
i
k i
n
19
92 and
ha
s b
ee
n
widely
us
e
d
t
o
s
olv
e
patte
r
n
rec
ogniti
on
pro
blem
s
,
par
ti
cl
e
identific
at
ion
prob
le
m
s
,
an
d
op
ti
m
i
zat
ion
pro
blem
s
[
5]. Ex
am
ples o
f
SV
M regressi
on
techn
iq
ues
inc
lud
e the f
ine S
VM, m
ediu
m
SV
M, coa
rse
-
ga
us
sia
n
SV
M,
a
nd
li
ne
ar
s
upport
ve
ct
or
m
achin
e
(LSVM
)
re
gressi
on
te
ch
ni
qu
e
[
13
]
.
The
LSV
M
regre
ssion
te
chn
iq
ue
ai
m
s
at
fin
di
n
g
a
li
near
f
unct
ion
f
(x)
with
t
he
m
ini
m
al
no
rm
va
lue
(
β’β
)
t
hat
m
akes
the
f
un
ct
ion
f(x)
t
o
be
as
fl
at
as
possible
us
in
g
wh
at
is
ca
ll
ed
pri
m
a
l
fo
rm
ula
an
d
dual
form
ula.
Th
e
pr
im
al
fo
rm
ula
is
br
ie
fly
desc
rib
e
d
us
in
g
Eq
ua
ti
on
s
(1
-
2)
,
w
hile
E
quat
io
ns
(
3
-
6)
il
lustrate
the
w
orkin
g
pr
i
nciple
for
t
he
du
al
form
ula u
sed
in
li
nea
r
S
VM
r
egr
es
sio
n
a
naly
sis
,
f
(
x
)
x
b
(
1
)
N
*
nn
n1
1
J
(
)
C
(
2
(
2
)
NN
*
*
'
i
i
j
j
i
j
i
1
j
1
1
L(
)
(
)
(
)
X
X
V
2
(
3
)
NN
**
i
i
i
i
i
i
1
i
1
V
(
)
y
(
)
(
4
)
N
*
n
n
n
n1
(
)
x
.
(
5
)
N
*'
n
n
n
n1
f
(
x
)
(
)
(
x
x
)
b
.
(
6
)
Wh
e
re
xn
a
re
the
data
set
s
c
om
pr
isi
ng
of
N
obse
rv
at
io
ns,
yn
is
the
re
sp
onse
,
εn
an
d
εn
*
de
note
th
e
sla
ck
var
ia
bles for
e
ach
po
i
nt, whil
e αn a
nd αn
*
a
re
non
-
ne
gativ
e m
ult
ipli
ers
f
or eac
h ob
se
r
vation x
n
[14
]
.
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.
10
, N
o.
3
,
June
2018
:
1080
–
1089
1082
2.2.
Ratio
na
l
Qu
ad
r
at
ic
-
G
au
ssi
an
P
roce
ss R
e
gressi
on
Ra
ti
on
al
quad
r
at
ic
gau
ssia
n
proces
s
re
gr
es
sion
(RQ
GP
R
)
is
a
ty
pe
of
ga
us
sia
n
p
ro
ces
s
regressi
on
(G
PR
)
te
ch
nique
t
hat
us
es
rat
ion
al
q
ua
dr
at
ic
kernel
(c
ovari
ance)
i
n
s
olv
in
g
opti
m
iz
a
ti
on
,
regressio
n,
t
ra
i
nin
g
and
pr
e
dicti
on
prob
le
m
s
[1
5
]
.
The
re
gr
ess
ion
kernel
hel
ps
to
de
te
rm
i
ne
the
cha
ract
erist
ic
s
(e.g.
f
it
ness,
sm
oo
thn
ess
,
pe
rio
dici
ty
,
et
c.)
of
a
f
un
ct
io
n
f
(
x)
.
T
he
R
QGPR
ke
r
nel
is
denote
d
usi
ng
E
quat
io
n
(
7).
T
he
RQGP
R
te
c
hniqu
e
e
xh
i
bits
so
m
e
si
m
il
arities
with
the
s
qu
a
re
d
ex
pone
ntial
GP
R
w
he
n
the
scal
e
m
ixtur
e
par
am
et
er
(σ)
f
ro
m
E
qu
at
io
n
(
7)
a
ppr
oach
e
s
zero
[16
]
.
T
he
above
-
m
entioned
s
qua
re
ex
po
nen
ti
al
GP
R
k
ern
el
can
be rep
rese
nted usi
ng E
qu
at
ion
(8),
2
2
rq
2
(
x
x
)
K
(
x
,
x
)
[
1
]
2l
(
7
)
2
2
se
2
(
x
x
)
K
(
x
,
x
)
.
e
x
p
[
]
2l
(
8
)
Wh
e
re
K
(
x,
x’)
is
the
co
va
riance
ke
rn
el
,
va
riable
x
an
d
x’
a
re
the
i
nput
var
ia
bles,
σ
2
is
the
m
a
xim
u
m
cov
a
riance
,
a
nd
l
is
the scal
ed
-
le
ng
t
h
use
d
to d
et
erm
ine
how
qu
ic
kly
a
G
PR
va
ries
with
the
in
put
va
riable
(
x)
.
Othe
r
e
xam
ple
s of
GP
R
kerne
ls i
nclu
de
m
at
e
rn s/2 ke
rn
el
,
a
nd exp
on
e
ntial
G
PR
ke
rn
el
[
17
]
-
[18
]
.
2.3.
Art
ific
ial
N
eur
al
Net
w
ork (AN
N)
ANN
is
a
m
ac
hin
e
le
ar
ning
te
chn
i
qu
e
th
at
works
li
ke
the
hu
m
an
br
ai
n
and
is
us
e
d
to
so
lve
bot
h
li
near
an
d
non
-
li
near
ta
s
ks
.
ANN
c
om
pr
ise
s
of
t
hr
ee
la
y
ers;
the
in
put
la
ye
r,
hidde
n
l
ay
er,
an
d
the
ou
t
pu
t
la
ye
r
[1
9
]
.
T
he
inp
ut
la
ye
r
re
cei
ves
the
inf
orm
ation
(traini
ng
data),
proce
sses
the
data
thr
ough
le
arn
i
ng,
an
d
giv
es
ou
t
pr
e
di
ct
ed
outp
ut
dat
a
thr
ough
the
outp
ut
la
ye
r.
Th
e
hidden
la
ye
r
is
an
inv
isi
ble
la
ye
r
with
it
s
outp
ut
interco
nnect
ed
to
the
in
puts
of
so
m
e
oth
e
r
ne
uro
ns
[
20
]
.
I
n
photov
oltai
c
syst
e
m
s,
AN
N
input
va
riables can be
the irr
adia
nce (G)
,
te
m
per
at
ure (T)
, ope
n
ci
r
cuit vo
lt
age (V
oc
)
w
hile t
he
ANN
predict
e
d respon
se ca
n
be
duty
cy
cl
e
(D
),
pr
e
di
ct
ed
current,
predict
ed
vo
lt
a
ge
,
or
pr
e
dicte
d
PV
po
wer.
Th
e
ANN
ne
uro
ns
process,
e
val
uate
the
input
sig
na
l
us
ing
li
nea
r
m
et
ho
d,
t
hen
c
om
par
e
with
it
s
su
m
by
m
ean
s
of
a
non
-
li
ne
ar
f
un
ct
io
n
known
as
act
ivati
on
f
unc
ti
on
,
a
nd
s
en
ds
the
co
ns
e
qu
e
nce
to
oth
e
r
ne
uro
ns
.
ANN
has
tw
o
com
m
on
ty
pe
s,
the
f
eed
-
forw
a
r
d neural
n
et
w
ork
and t
h
e
recurre
nt
ne
ur
al
netw
ork [
21
].
A ne
uro
n
i
s m
od
el
le
d
us
i
ng equati
on
(9)
,
M
mm
m1
Z
X
W
(
9
)
wh
e
re
the
in
put
var
ia
bles
are d
en
oted
by
X
1
,
X
2
,
X
3
,
…,
X
m
an
d
the
res
pec
ti
ve
weigh
t
of
the
ind
i
vidual
inputs
are
denoted
by
W
1
,
W
2
,
W
3,
…,
W
m
res
pec
ti
vely
[5
]
,
[22
]
.
Ma
them
at
ic
a
l
ly
,
the
ne
uro
ns
in
the
hi
dden
la
ye
r
can
be
est
im
ated
us
in
g
e
quat
ion (
10),
io
he
(
N
N
)
NN
2
(
10
)
wh
e
re
N
h
is
th
e
hid
de
n
la
ye
r,
N
i
is
the
inpu
t
la
ye
r,
and
N
o
is
the
ou
t
pu
t
l
ay
er.
So
m
e
drawb
ac
ks
with
ANN
te
chn
iq
ue
i
nclud
e
l
onge
r
pr
oc
essing
ti
m
e
fo
r
la
r
ge
netw
orks
,
com
plexity
of
the
A
NN
al
gorithm
,
and
ANN
proce
dure
dem
ands
that
the
syst
e
m
is
first
t
raine
d
us
in
g
pri
or
data
set
s
and
the
c
ollec
ti
on
of
these
tra
inin
g
d
at
a sets m
igh
t be c
um
ber
so
m
e [
23
]
-
[25]
.
3.
SIMULATI
O
N MO
DEL
To
in
vestigat
e
the f
easi
bili
ty
with the use
of
r
eg
ressi
on
lea
r
ning a
lg
or
it
hm
s in
trac
k
in
g
th
e m
a
xi
m
u
m
powe
r
po
i
nt
in
a
sta
nd
-
al
one
phot
ovoltai
c
syst
e
m
,
an
exp
erim
ent
was
co
nducte
d
us
in
g
a
co
m
ple
te
photov
oltai
c
syst
e
m
that
inclu
de
s
a
so
lt
ech
1S
T
H
-
215
-
P
P
V
pa
nel,
m
od
ifie
d
c
uk
DC
-
D
C
conver
te
r,
MPP
T
con
t
ro
ll
er,
an
d
a
20
Ω
resist
ive
loa
d.
T
he
tr
ai
ning
da
ta
s
et
s
wer
e
c
olle
ct
ed
f
ro
m
PSIM
so
ft
war
e
.
T
able
1
il
lustrate
s
the
sp
eci
ficat
ions
of
the
us
e
d
PV
pan
el
an
d
the
DC
-
DC
conve
r
te
r
in
this
study
.
The
PV
ef
fi
ci
ency,
load
e
ff
ic
ie
ncy, and
DC
-
DC c
onve
rter loss
w
ere
ob
ta
ine
d us
ing
E
quat
io
ns
(
11
-
13),
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
Op
ti
miz
atio
n o
f PV Sy
ste
ms U
sing D
ata Mi
ni
ng an
d
Re
gr
es
sion Le
ar
ne
r…
(
Adeday
o
M
. F
arayol
a
)
1083
t
p
v
(
m
a
x
)
t
0
t
p
v
(
m
p
p
t
)
t
0
P
.
d
t
P
V
e
f
f
ic
ie
n
c
y
a
t
M
P
P
T
=
P
V
.
d
t
(
11
)
t
o
u
t
(
m
p
p
t
)
t
0
t
p
v
(
m
p
p
t
)
t
0
P
.
d
t
M
C
U
K
l
o
a
d
e
f
f
ic
ie
n
c
y
a
t
M
P
P
T
=
P
V
.
d
t
(
12
)
M
C
U
K
L
o
s
s
e
s
i
n
p
u
t
p
o
w
e
r
–
o
u
t
p
u
t
p
o
w
e
r
(
13
)
Wh
e
re
P
pv(
mppt
)
is
the
1S
T
H
-
215
-
P
rate
d
po
wer
at
STC
(st
and
a
r
d
te
st
con
diti
on)
,
P
pv(
ma
x)
is
the
PV
extracte
d
powe
r,
a
nd
P
out
is t
he o
utput p
ow
e
r
at
t
he 20
Ω resist
ive.
Table
1.
T
he
U
sed
P
V P
an
el
a
nd MC
U
K DC
-
DC
C
onve
rter
Sp
eci
ficat
ion
s
So
lar
Pan
el Spec
if
icatio
n
s
Mcuk
Sp
ecif
icatio
n
s
PV M
o
d
el
1
STH
-
215
-
P
L
1
4
m
H
Stan
d
ard Te
st Co
n
d
itio
n
1
0
0
0
W
/
m
2
,
25
°C
L
2
4
m
H
Maxi
m
u
m
Voltag
e
(
V
mo
)
2
9
.0V
C
1
1
0
0
µF
Maxi
m
u
m
curr
en
t
(I
mp
)
7
.35
A
C
2
1
0
0
µF
Maxi
m
u
m
Powe
r
(
P
mp
)
2
1
3
.15
W
R
0
2
0
Ω
N
s
-
n
u
m
b
e
r
o
f
cell
in series
60
C
0
2
7
0
µF
I
sc
-
sh
o
rt
cir
cu
it c
u
rr
en
t
7
.84
A
V
o
c
–
o
p
en
cir
cu
it vo
ltag
e
3
6
.30
V
Te
m
p
.
coef
f
icien
t of
I
sc
-
0
.36
0
9
9
% / °
C
Te
m
p
.
coef
f
icien
t of
V
oc
0
.10
2
% / °
C
A
-
Dio
d
e idealit
y
f
acto
r
0
.98
1
1
7
Figure
1
pr
ese
nts
the
flo
wchart
al
go
rithm
of
the
li
near
suppo
rt
vector
m
ac
hin
e
(L
SV
M
)
regress
i
on
te
chn
iq
ue.
T
he
LSV
M
m
od
el
li
ng
was
do
ne
in
three
fo
lds
.
The
first
fo
l
d
dealt
with
the
optim
iz
ation
an
d
gen
e
rati
on
of
a
fitness
f
un
ct
io
n
(yfit
)
usi
ng
LSV
M
re
gr
e
ss
ion
ke
r
nel
and
few
c
ollec
te
d
sa
m
ples
of
the
PSIM
data
set
s
(
19
instances
)
in
t
r
ai
n
ing
the
m
odel
.
The
data
se
ts
com
pr
ise
of
two
in
put
va
riables
(
dif
fer
e
nt
le
vels
of
ir
rad
ia
nce(G)
a
nd
te
m
per
at
ur
e
(T
))
as
pr
edict
or
s
(X)
an
d
one
outp
ut
va
riable
(r
e
fer
e
nce
cu
rr
e
nt
(Ir
ef*
)
)
as
the
respo
ns
e.
T
he
LSV
M
re
gressi
on
kernel
was
then
us
ed
to
pr
e
dict
the
respon
ses
(Iref
)
fo
r
an
a
dd
it
io
nal
110
instances
c
omprisin
g
of
va
ri
ables
G
an
d
T.
The
seco
nd
f
ol
d
dealt
with
th
e
trai
ning,
te
sti
ng,
a
nd
validat
ion
of
the
LS
VM
ne
wly
pr
e
dicte
d
data
set
s
us
i
ng
ANN
te
c
hn
i
que.
T
he
L
SV
M
gen
e
rated
dat
a
s
et
s
wer
e
sp
l
it
in
the
pro
portion 7
5%
for t
rainin
g,
15% test
ing, a
nd 15%
v
al
idat
ion
.
Fo
r
the
thi
rd
f
old
,
Fi
gure
2
di
sp
la
ys
a
co
m
plete
sta
nd
-
al
on
e
PV
syst
e
m
d
esi
gn
e
d
us
i
ng
ANN
-
LS
VM
te
chn
iq
ue.
T
he
ANN
-
L
SV
M
ou
t
pu
t
(
ref
e
re
nc
e
cur
re
nt
(
Ir
e
f*))
w
as
com
par
ed
with
the
PV
cu
rr
e
nt
(Ip
v)
as
error
sig
nal
(
Iref*
–
I
pv).
T
he
erro
r
si
gn
al
w
as
pas
sed
th
rough
a
power
-
in
te
gr
al
(P
I
)
c
ontrolle
r
f
or
fin
e
t
un
i
ng
and
outp
uts
the
du
ty
cy
cl
e
sig
nal
(
D)
.
The
duty
cy
cl
e
sign
a
l
was
tra
ns
m
itt
ed
th
rou
gh
a
pulse
wi
dth
m
od
ulat
or
(PW
M
)
as
p
ulse
sig
nal
that
was
us
e
d
to
ac
ti
vate
the
Mos
fet
gate
of
t
he
DC
-
DC
co
nve
rter.
Ta
ble
2
di
sp
la
ys
the
aver
a
ge
te
sti
ng
er
ror
of
the
trai
ned
A
NN
-
LS
VM
MPPT
te
ch
nique
us
ing
a
total
of
12
9
sam
ple
s
that
com
pr
ise
of
10
sam
ples
fr
om
the
PS
IM
data
set
s
and
119
sa
m
ples
fr
om
the
gen
e
rated
A
N
N
-
L
SV
M
data
set
s
in the p
rop
or
ti
on 70%
trainin
g,
15% test
in
g, an
d 15%
valid
at
ion
.
Fo
r
the
rati
on
al
qu
a
drat
ic
ga
us
sia
n
pr
oces
s
re
gr
es
sio
n
(
RQGP
R
)
te
c
hniq
ue,
sim
il
ar
proce
dures
as
sh
ow
n
in
Fig
ur
e
1,
a
nd
a
blo
ck
diag
ram
a
s
show
n
in
Fi
gu
re
2
of
a
com
plete
PV
syst
e
m
bu
t
us
in
g
R
QGPR
al
gorithm
and
ANN
-
R
QGP
R
MPPT
te
chn
iq
ue
we
re
use
d.
Ta
ble
3
presents
the
A
NN
-
RQ
GP
R
trai
ning
,
te
sti
ng
,
a
nd
va
li
dation
sta
ti
sti
cs.
Wh
ere
a
va
lue
of
R
that
is
appro
ac
hing
1.000
00
a
nd
a
MSE
appr
oa
chin
g
0.000
00 v
al
ida
t
e a w
el
l t
raine
d
m
od
el
.
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.
10
, N
o.
3
,
June
2018
:
1080
–
1089
1084
S
t
a
r
t
R
e
g
r
e
s
s
i
o
n
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
m
U
s
e
t
h
e
n
e
w
l
y
p
r
e
d
i
c
t
e
d
s
a
m
p
l
e
s
(
i
n
p
u
t
s
G
a
n
d
T
,
o
u
t
p
u
t
I
r
e
f
f
o
r
A
N
N
t
r
a
i
n
i
n
g
,
t
e
s
t
i
n
g
a
n
d
v
a
l
i
d
a
t
i
o
n
)
(
1
9
+
1
1
0
s
a
m
p
l
e
s
)
U
s
e
t
he
ge
ne
ra
t
e
d fi
t
ne
s
s
func
t
i
on
t
o pre
di
c
t
ne
w
t
ra
i
ni
ng i
ns
t
a
nc
e
s
(110
s
a
m
pl
e
s
)
S
t
op a
t
t
=
0.
4
s
Im
port
f
e
w
P
S
IM
t
ra
i
ni
ng
da
t
a
(G
, T
, I
pv
) for pre
di
c
t
i
on
(1
9
s
a
m
pl
e
s
)
G
e
ne
ra
t
e
F
i
t
ne
s
s
func
t
i
on
y
f
i
t
=
t
r
a
i
n
e
d
C
l
a
s
s
i
f
i
e
r
.
p
r
e
d
i
c
t
F
c
n
(
X
)
us
i
ng
Re
gre
s
s
i
on
l
e
a
rni
ng
K
e
rn
e
l
s
W
he
re
X
a
re
t
he
i
n
put
pre
di
c
t
ors
(G
a
nd
T
) a
nd
yfi
t
i
s
t
he
ou
t
put
Re
s
pons
e
(Ire
f)
E
rror( e
) =
Ipv
–
Ire
f*
E
r
r
or
e
i
s
p
a
s
s
e
d
t
o
P
ID
c
ont
rol
l
e
r fo
r
t
uni
ng
a
nd
t
o o
bt
a
i
n
D
ut
y
c
yc
l
e
,
D
D
ut
y Cyc
l
e
D
i
s
pa
s
s
e
d t
o P
W
M
c
ont
ro
l
l
e
r a
t
50K
H
Z
fre
que
nc
y a
s
pu
l
s
e
T
o M
os
fe
t
of
D
C
-
D
C
Con
ve
rt
e
r
U
s
e
t
he
A
N
N
s
ys
t
e
m
t
o obt
a
i
n out
put
Ire
f*
Figure
1. The
ANN
-
LS
VM a
lgorit
hm
Figure
2. Com
plete
PV syst
em
d
esi
gn
ed
u
si
ng AN
N
-
L
SVM
MPPT tec
hniq
ue
Table
2.
T
he
R
egr
es
sio
n
(R
)
a
nd Mea
n S
qu
a
r
e Er
ror (M
SE)
S
ta
ti
sti
cs
fo
r A
NN
-
LS
VM T
e
chn
i
qu
e
ANN
-
LSV
M
Sa
m
p
les
MSE
Reg
ressio
n
(
R)
Tr
ain
in
g
91
1
.94
2
4
7
e
-
9
9
.99
9
9
9
e
-
1
Testin
g
19
1
.26
6
0
0
e
-
8
9
.99
9
9
9
e
-
1
Valid
atio
n
19
6
.20
5
2
3
e
-
7
9
.99
9
9
9
e
-
1
Table
3.
ANN
-
R
Q
GP
R
R
egr
essi
on (
R
)
and
Me
a
n
S
qu
are E
rror
(MS
E)
S
ta
ti
sti
cs
ANN
-
RQ
GPR
Sa
m
p
les
MSE
Reg
ressio
n
(
R)
Tr
ain
in
g
91
1
.28
4
8
2
e
-
9
9
.99
9
9
9
e
-
1
Testin
g
19
5
.33
0
8
0
e
-
9
9
.99
9
9
9
e
-
1
Valid
atio
n
19
1
.79
4
9
4
e
-
7
9
.99
9
9
9
e
-
1
Fo
r
t
he
c
onve
ntion
al
ANN
t
echn
i
qu
e
,
F
i
gure
3
s
how
s
th
e
com
plete
PV
syst
em
desi
gn
e
d
us
in
g
a
conve
ntion
al
ANN
MPPT
te
chn
i
qu
e
.
U
nlike
the
ANN
-
LSV
M
an
d
A
NN
-
RQ
GP
R
te
chn
i
qu
e
s
that
were
trai
ned
us
in
g
optim
iz
ed
data
set
s
from
the
r
e
gr
es
sio
n
le
arni
ng
pr
e
dicti
on
s
,
t
he
co
nventio
nal
A
NN
al
go
r
it
h
m
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
Op
ti
miz
atio
n o
f PV Sy
ste
ms U
sing D
ata Mi
ni
ng an
d
Re
gr
es
sion Le
ar
ne
r…
(
Adeday
o
M
. F
arayol
a
)
1085
was
trai
ned
us
i
ng
real
data
se
ts
(12
9
insta
nc
es
of
G
,
T
,
an
d
I
ref)
t
hat
we
re
colle
ct
ed
en
ti
rely
fr
om
the
PSIM
so
ft
war
e
t
hro
ugh
a
dy
nam
ic
PV
sim
ulati
on
.
Table
4
dis
play
s
the
erro
r
s
ta
ti
sti
cs
fo
r
th
e
co
nv
e
ntio
nal
ANN
trai
ning,
te
sti
ng, a
nd
validat
ion.
Table
4.
A
NN
R
egr
essi
on (
R
)
and
Me
a
n
S
qu
are E
rror
(MS
E)
S
ta
ti
sti
cs
Co
n
v
en
tio
n
al ANN
Sa
m
p
les
MSE
Reg
ressio
n
(
R)
Tr
ain
in
g
91
1
.06
3
0
4
e
-
5
9
.99
9
9
8
e
-
1
Testin
g
19
1
.05
6
6
3
e
-
5
9
.99
9
9
3
e
-
1
Valid
atio
n
19
7
.10
0
1
9
e
-
5
9
.99
9
9
0
e
-
1
Figure
3. Com
plete
PV syst
em
d
esi
gn
ed
u
si
ng con
ven
ti
ona
l ANN tec
hn
i
que
4.
RESU
LT
S
A
ND AN
ALYSIS
Table
5
a
nd
F
igures
4
-
9
pr
e
sent
the
ta
bula
te
d
res
ults
a
nd
the
gr
a
phic
al
resu
lt
s
for
t
he
co
nduct
e
d
exp
e
rim
ent
us
ing
c
onve
ntio
na
l
ANN
te
ch
ni
qu
e
,
no
n
-
c
onve
ntion
al
A
NN
-
LSV
M
a
nd
no
n
-
c
onve
ntio
nal
ANN
-
RQGP
R
MP
P
T
te
chn
i
qu
e
unde
r
th
ree
di
fferent
weather
conditi
ons
(
N
OCT,
PTC
,
an
d
STC)
.
T
he
NO
CT
is
the
norm
al
op
erati
ng
ce
ll
te
m
per
at
ur
e
wh
e
re
the
irra
diance
(G)
is
800
W/
m
2
and
the
am
bient
te
m
per
at
ur
e
(T)
is
47
.
40
°C
.
S
TC
is
the
sta
nd
ar
d
te
st
cond
it
ion
wh
e
re
G
is
10
00
W
/m
2
and
T
is
25
°
C
wh
il
e
PTC
is
the
PVUS
A
te
st co
nd
it
io
n wh
e
re
G
is
1000
W/m
2
a
nd T is
20 °
C
Fo
r
case
1
(NO
CT)
,
where
G
is
800
W
m
-
2
a
nd
T
is
47.
40
°C
,
obta
ined
resu
lt
s
show
that
the
conve
ntion
al
ANN
an
d
the
non
-
co
nventio
nal
ANN
-
R
Q
G
PR
te
chn
iq
ue
had
ti
e
(sam
e)
and
best
res
ults
at
bo
th
the
PV
en
d
a
nd
at
the
20
Ω
resist
ive
-
loa
d
end
(
73.09%
PV
e
ff
ic
ie
ncy
and
69.
79%
outp
ut
loa
d
ef
fici
ency
)
wh
il
e
ANN
-
L
SV
M
disp
la
ye
d
the
lo
west
perform
ance
(
72.67%
PV
ef
fici
ency,
69.43%
loa
d
e
ff
ic
ie
ncy).
Howe
ver,
the
ANN
-
LS
VM
exh
i
bited
the
lowest
DC
-
D
C
con
ve
rter
powe
r
loss
un
de
r
NO
CT
c
onditi
on.
Figures
4
-
5
di
sp
la
y
the
grap
hical
res
ults
of
the
e
xtracte
d
PV
powe
r
a
nd
the
outp
ut
loa
d
powe
r
from
the
P
V
syst
e
m
us
ing
conve
ntion
al
ANN,
non
-
co
nv
e
ntio
nal
A
NN
-
LS
VM,
A
NN
-
RQ
GP
R
t
echn
i
qu
e
,
an
d
under
NO
CT
w
eat
her co
nd
it
io
n.
Fo
r
ca
se
2
(
P
TC),
w
he
re
G
is
1000
W
m
-
2
an
d
T
is
20
°C,
A
NN
-
RQ
GP
R
ha
d
the
best
res
ult
(10
1.95
%
P
V
eff
ic
ie
ncy
a
nd
97.
71%
r
esi
sti
ve
load
eff
ic
ie
ncy)
,
w
hile
A
NN
-
LS
VM
un
derper
form
ed
(10
0.40
%
P
V
eff
ic
ie
ncy
an
d
97
.
71%
loa
d
eff
ic
ie
ncy)
.
Sim
il
arly
,
the
D
C
-
DC
co
nvert
er
loss
in
po
w
er
with
ANN
-
LS
VM
r
egr
es
sio
n
te
ch
nique
was
the
lowest
(8.56
W).
Fig
ures
6
-
7
disp
la
y
the
grap
hical
res
ults
of
th
e
extracte
d
P
V
powe
r
an
d
the outp
ut
load
pow
er
from
the
PV
syst
e
m
us
ing
ANN,
A
N
N
-
L
SV
M, ANN
-
R
QGPR
te
chn
iq
ue, a
nd
unde
r
PTC
w
e
at
her
c
onditi
on
.
Fo
r
case
3
(
ST
C),
where
G
is
1000
W
m
-
2
a
nd
T
is
25
°C,
both
A
N
N
a
nd
ANN
-
RQ
GPR
achiev
e
d
cl
os
e
res
ults
at
the
P
V
e
nd
a
s
eq
ual
powe
r
s
and
ef
fici
encies
were
ext
racted
f
r
om
the
PV
syst
em
(2
12
.90
W
PV
pow
e
r
a
nd 9
9.8
8%
PV
ef
f
ic
ie
ncy)
w
he
re
as
the PV
pa
ne
l
did
und
e
r
perf
or
m
us
in
g
ANN
-
L
SV
M (
209.9
0
W
PV
i
nput
P
V
powe
r
a
nd
98.
48%
P
V
e
f
f
ic
ie
ncy).
How
ever,
at
the
20
Ω
resist
ive
en
d,
A
NN
-
R
QGPR
ov
e
r
perform
ed
by
e
xtracti
ng
the
m
axi
m
u
m
ou
t
pu
t
powe
r
a
nd
ef
fici
ency
(
204.1
6
W
resis
ti
ve
-
loa
d
powe
r
a
nd
95.78%
loa
d
e
ff
ic
ie
ncy)
.
Als
o,
the
A
N
N
-
L
SV
M
MPPT
te
chn
i
qu
e
unde
rp
e
rfor
m
ed
as
201.4
5
W
po
w
er
was
pro
du
ce
d
at
the
resist
ive
load
en
d
an
d
with
a
load
eff
ic
i
ency
of
94.
45
%.
Figures
8
-
9
disp
la
y
the
graph
ic
al
resu
lt
s
of
the
e
xtracted
P
V
power
an
d
t
he
ou
tpu
t
l
oad
po
we
r
f
r
om
the
PV
syst
e
m
us
ing
ANN,
A
N
N
-
L
SV
M,
ANN
-
R
Q
GP
R
te
chn
iq
ue, a
nd
unde
r
PTC
w
e
at
her
c
ondit
ion.
In
a
ddit
ion
,
at
STC,
the
ANN
-
L
SV
M
had
the
lowe
st
DC
-
DC
c
onve
rter
powe
r
loss
as
8.45
W
was
dissipated
at
the
DC
-
DC
c
onve
rter
w
hile
powe
r
dissi
pated
wit
h
A
N
N
-
RQGP
R
M
PP
T
te
chn
i
que
in
al
l
the
three
en
vir
onm
ental
con
diti
on
cases
(
N
O
CT,
PTC,
and
STC)
wer
e
t
he
highest.
Al
so
,
f
ro
m
the
a
naly
se
d
regressio
n
(R)
and
the
m
ean
-
squa
re
-
e
rror
(
MSE)
re
su
lt
s
sh
ow
n
in
Tabl
es
2
-
4,
A
N
N
-
RQGP
R
ha
d
t
he
be
st
resu
lt
wh
il
e
A
NN h
a
d
t
he worst t
raini
ng r
es
ult.
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.
10
, N
o.
3
,
June
2018
:
1080
–
1089
1086
Table
5.
Res
ults o
f
the
con
du
c
te
d
ex
pe
rim
ent
und
e
r varie
d weat
he
r
c
onditi
ons
W
eath
er
con
d
itio
n
s
VALU
ES
ANN
ANN
-
LSV
M
ANN
-
RQ
GPR
G =
8
0
0
W
m
-
2
T
= 47
.40
°C
(NOCT
)
CASE 1
PV curr
en
t (
A
)
5
.80
0
5
.71
5
5
.80
2
PV vo
ltag
e (
V
)
2
6
.57
2
7
.04
2
6
.77
Load
curre
n
t
(A)
-
2
.72
7
-
2
.72
0
-
2
.72
7
Load
vo
ltag
e (
V)
-
5
4
.55
-
5
4
.41
-
5
4
.55
PV
p
o
wer (
W
)
1
5
5
.80
1
5
4
.90
1
5
5
.80
Load
po
wer
(
W
)
1
4
8
.76
1
4
8
.00
1
4
8
.76
DC
-
DC lo
ss
es (
W
)
7
.04
6
.90
7
.04
PV E
f
f
icien
cy
(%)
7
3
.09
7
2
.67
7
3
.09
Load
ef
f
icien
cy
(
%)
6
9
.79
6
9
.43
6
9
.79
G =
1
0
0
0
W
m
-
2
T
= 20
°
C
(PT
C)
CASE 2
PV curr
en
t (
A
)
7
.14
1
6
.92
3
7
.15
9
PV vo
ltag
e (
V
)
3
0
.30
3
0
.86
3
0
.25
Load
curre
n
t
(A)
-
3
.22
6
-
3
.20
5
-
3
.22
7
Load
vo
ltag
e (
V)
-
6
4
.52
-
6
4
.10
-
6
4
.54
PV po
wer (
W
)
2
1
7
.00
2
1
4
.00
2
1
7
.30
Load
po
wer
(
W
)
2
0
8
.14
2
0
5
.44
2
0
8
.27
DC
-
DC lo
ss
es (
W
)
8
.86
8
.56
9
.03
PV E
f
f
icien
cy
(%)
1
0
1
.81
1
0
0
.40
1
0
1
.95
Load
ef
f
icien
cy
(
%)
9
7
.65
9
6
.38
9
7
.71
G =
1
0
0
0
W
m
-
2
T
= 25
°
C
(ST
C)
CASE 3
PV curr
en
t (
A
)
7
.17
4
6
.94
5
7
.17
9
PV vo
ltag
e (
V
)
2
9
.58
3
0
.16
2
9
.57
Load
curre
n
t
(A)
-
3
.19
4
-
3
.17
4
-
3
.19
5
Load
vo
ltag
e (
V)
-
6
3
.89
-
6
3
.47
-
6
3
.90
PV po
wer (
W
)
2
1
2
.90
2
0
9
.90
2
1
2
.90
Load
po
wer
(
W
)
2
0
4
.06
2
0
1
.45
2
0
4
.16
DC
-
DC lo
ss
es (
W
)
8
.84
8
.45
8
.74
PV E
f
f
icien
cy
(%)
9
9
.88
9
8
.48
9
9
.88
Load
ef
f
icien
cy
(
%)
9
5
.74
9
4
.51
9
5
.78
Figure
4. G
raph
of 1
S
TH
-
215
-
P i
nput
power at
NOCT
Figure
5. G
raph
of
1STH
-
215
-
P
ou
t
put p
owe
r
at
NO
CT
Figure
6. G
raph
of 1
S
TH
-
215
-
P i
nput
power at
PTC
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
Op
ti
miz
atio
n o
f PV Sy
ste
ms U
sing D
ata Mi
ni
ng an
d
Re
gr
es
sion Le
ar
ne
r…
(
Adeday
o
M
. F
arayol
a
)
1087
Figure
7. G
raph
of 1
S
TH
-
215
-
P
ou
t
put p
owe
r
at
PTC
Figure
8. G
raph
of 1
S
TH
-
215
-
P i
nput
power at
STC
Figure
9. G
raph
of 1
S
TH
-
215
-
P
ou
t
put p
owe
r
at
STC
5.
CONCL
US
I
O
N
S
This
pap
e
r
pr
e
sents
a
n
in
nova
ti
ve
us
e
of
a
par
ti
cula
r
ty
pe
of
data
m
ining
te
ch
nique
kn
own
as
the
rati
on
al
qu
a
dr
at
ic
gau
ssia
n
process
regres
sion
(RQ
GP
R)
le
arn
in
g
al
go
rithm
to
track
the
m
axi
m
u
m
powe
r
po
i
nt
in
PV
sy
stem
s.
Find
in
gs
su
ggest
that
op
ti
m
iz
ation
of
PV
syst
em
s
us
ing
R
QG
PR
t
echn
i
qu
e
ca
n
be
us
ed
to
ext
ract
m
axi
m
u
m
po
we
r
f
ro
m
a
phot
ovol
ta
ic
pan
el
unde
r
diff
e
ren
t
we
at
her
c
on
diti
on
s.
T
he
R
QGP
R
can
su
ccess
fu
ll
y
ge
ner
at
e
the
la
r
ge,
acce
ptable
and
acc
ur
at
e
trai
ning
data
s
et
s
need
e
d
to
trai
n
the
super
vised
m
achine
le
arn
i
ng
te
ch
niques
l
ike
ANN
an
d
ANFIS
for
MP
PT
ta
sk
s.
Als
o,
the
m
ean
-
squa
re
-
e
rror
(M
SE)
and
regressio
n
(R
)
error
sta
t
ist
ic
s
with
RQ
GP
R
te
chn
i
qu
e
wer
e
bette
r
than
t
ha
t
of
co
nventio
nal
A
NN
a
nd
l
inear
su
pp
or
t
vecto
r
m
achine
(
LSVM
)
re
gr
e
ssio
n
te
chn
iq
ues
.
Re
su
lt
s
co
nfi
rm
e
d
that
R
Q
GP
R
te
chn
i
qu
e
ex
hi
bited
an
im
pr
ov
e
d
trackin
g
powe
r
an
d
e
ff
ic
ie
nc
y
com
par
ed
to
the
li
nea
r
su
pp
or
t
vecto
r
m
achine
(LSVM
)
regressio
n
te
c
hniq
ue.
REFERE
NCE
S
[1]
Y.
Jin
y
ue
,
Hand
book
of
Cl
ea
n
E
ner
g
y
S
y
s
te
m
s, 6 Volum
e
Set
,
V
olume
5,
6th ed.,
John W
il
e
y
&
S
ons,
2015.
[2]
A.
M.
Fara
y
ol
a,
A.
N.
Hasan,
an
d
A.
Ali,
"Com
par
ison
of
m
odif
ie
d
in
cre
m
ental
conduc
t
anc
e
and
fuz
z
y
logi
c
m
pp
t
al
gorit
hm
using
m
odifi
ed
cuk
co
nver
te
r
,
"
in
8th
IEE
E
Int
ernati
on
al
Re
newab
le
E
nergy
Congress
(
IRE
C)
,
A
mman,
Jordan,
2017
.
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.
10
, N
o.
3
,
June
2018
:
1080
–
1089
1088
[3]
A.
M.
Fara
y
ola,
A.
N.
Hasan,
a
nd
A.
Ali
,
"Cur
ve
fit
t
ing
pol
y
no
m
ia
l
technique
c
om
par
ed
to
AN
FIS
te
chni
que
fo
r
m
axi
m
um
po
wer
point
tra
ck
ing
,
"
in
8th
IEE
E
In
te
rnational
Re
ne
wable
Ene
rgy
C
ongress
(
IRE
C)
,
Amman,
Jordan
,
2017.
[4]
A.
M.
Far
a
y
ola,
A.
N.
Hasan,
a
nd
A.
Ali
,
"Im
pl
ementa
t
ion
of
m
odifi
ed
inc
r
eme
nta
l
conducta
n
c
e
and
fuz
z
y
log
i
c
m
ppt
te
chn
ique
s
using
m
cuk
con
ver
te
r
under
var
i
ous
envi
ronm
ental
condi
t
ions
,
"
A
ppli
ed
Solar
En
e
rgy
,
vol
.
53
,
no
.
2,
pp
.
1
-
13
,
201
7.
[5]
A.
M.
Fara
y
o
la,
"Com
par
at
ive
st
ud
y
of
diffe
r
ent
photovol
taic
MP
PT
te
chn
ique
s
u
nder
var
ious
wea
the
r
conditions
(The
sis
subm
it
te
d
in
par
t
ia
l
fu
lfi
ll
m
ent
for
th
e
degr
e
e
of
Master
s
in
Elec
t
ri
ca
l
Eng
ineeri
ng
),
"
Unive
rs
it
y
o
f
Johanne
sbur
g
,
J
ohanne
sburg,
South Afri
c
a, 2017
.
[6]
A.
M.
Fara
y
ol
a,
A.
N
.
Hasa
n,
A.
Ali
,
and
B.
Twa
la,
"D
istri
bu
ti
v
e
MP
PT
appr
oa
ch
usi
ng
AN
FIS
and
Perturb&
Obs
erv
e
technique
s
un
der
uniform
an
d
par
tial
shadin
g
condi
ti
ons
,
"
i
n
Inte
rnational
Confe
renc
e
on
Arti
ficial
Intelligence
and
Ev
o
l
uti
onary
Computati
ons
in
Eng
in
ee
ring
Syst
ems
&
P
ower,
Circui
t
and
Informati
on
Technol
ogi
es
(
ICPCIT
-
2017)
,
India
,
2017.
[7]
S.
R.
Nandurkar
and
M.
Rajee
v
,
“
Modeli
ng
si
m
ula
ti
on
&
d
esign
of
photovoltaic
arr
a
y
with
MP
PT
cont
rol
te
chn
ique
s
,”
Int
e
rnational
Journal
of
Applied
Pow
er
Engi
n
ee
ring
(
IJA
P
E)
,
vol. 3
,
n
o.
1
,
pp
.
41
-
50
,
2014.
[8]
J.
Cunli
ang
,
W
.
Yanxiong
,
and
W
.
Ze
rong
,
“
Photovolt
a
ic
arr
a
y
m
axi
m
um
powe
r
point
tr
ac
k
ing
base
d
on
improved
m
et
hod
,
”
TEL
KOMNIKA
,
vol
.
1
4,
no
.
2
,
pp
.
404
-
410,
2016
.
[9
]
A.
Jus
o
h,
R.
Ali
k
,
T
.
K
Guan
,
T
.
Sutikno
,
“
MP
P
T
for
PV
s
y
stem
base
d
on
var
ia
b
l
e
step
siz
e
p&o
al
gorit
hm
,
”
TEL
KOMNIKA
,
vol.
15,
no.
1
,
pp
.
79
-
92,
2017
.
[10]
S.
Doraha
ki,
“
A
surve
y
on
m
axim
um
po
wer
point
tra
ck
ing
m
et
h
ods
in
photovol
t
ai
c
power
s
y
s
tem
s
,”
Bul
l
et
i
n
of
El
e
ct
rica
l
Eng
in
ee
ring a
nd
Infor
matic
s
,
vo
l. 4,
n
o.
3
,
pp
.
169
-
17
5
,
2015
.
[11]
R.
J.
Prasanth
and
B.
T.
Sudh
aka
r,
"A
comprehe
nsive
r
eview
on
solar
PV
m
axi
m
um
powe
r
point
tr
ac
k
ing
te
chn
ique
s,"
Re
n
ewabl
e
and
Sust
ainabl
e
Ene
rgy
,
ELSEV
IE
R
,
vol.
67,
no
.
3
,
pp
.
82
6
-
847,
2017
.
[12
]
A.
H.
Nure
,
"D
i
stribut
ed
d
ata
m
ini
ng
using
m
ulti
ag
ent
d
ata,
"
In
te
rnational
R
ese
arch
Journal
o
f
Engi
ne
ering
an
d
Technol
ogy
(
IRJET)
,
vol.
4
,
no
.
3
,
pp
.
463
-
468
,
2
017.
[13
]
Ahm
e
d
Ali,
Ali
N.
Hasan
and
Tshil
idzi
Marwal
a
,
“
Perturb
and
O
bserve
base
d
on
fuz
z
y
logic
controlle
r
m
axi
m
um
power
point
trac
king
(MP
PT)”,
i
n
IEE
E
Inte
rnat
i
onal
Confe
ren
ce on
Re
n
ewabl
e
E
nergy
R
ese
arch
and
Appl
i
cat
ions
(
ICRE
RA
)
2014,
Mil
wau
ke
e
USA
,
2014
.
[14
]
P.
Dom
ingos,
T
he
Master
Algorit
hm
:
How
the
quest
for
the
ultim
at
e
learni
ng
m
ac
hine
will
re
m
ake
our
world
,
Unite
d
St
at
es:
B
asic
Boo
ks
,
2015
.
[15
]
A.
O.
Khedr
-
Ibr
ahi
m
,
Regre
ss
io
n
base
d
m
ult
i
-
st
age
a
lgori
thm
to
tra
ck
the
m
axim
um
po
wer
point
for
photovolt
ai
c
s
y
stems
(in
fu
lfi
l
lment
of the
Master
of
Sci
enc
e
),
W
at
erl
oo,
Ontar
i
o,
Ca
n
ada:
Uni
v
ersity
o
f
Wat
erlo
o
,
2012
.
[16
]
C.
E
Rasm
uss
en
and
C.
K.
I.
W
illia
m
s
,
Gauss
ia
n
proc
esses
for
ma
chi
n
e
learni
ng
,
Cambridge,
Mas
sac
husett
s
:
MIT
Press
,
2006.
[17
]
A.
W
il
sons
and
R.
Adam
s,
"G
aussian
proc
ess
ke
rne
ls
for
p
at
t
ern
discove
r
y
and
e
xtra
p
ol
at
ion
,
"
In
Proceedi
ngs
of
the
30
th
Int
ernat
ional
Conf
ere
nc
e
on
Ma
chi
n
e
Le
arning
(
ICML
-
1
3)
,
Atla
nt
a, USA
,
2013
.
[18
]
A.
G.
W
il
son,
H.
Zhi
ti
ng
,
S.
Ruslan,
and
E
.
P.
Xing,
"D
ee
p
ker
ne
l
le
arn
ing
,
"
in
Art
if
icial
Int
el
l
ige
n
ce
and
Stat
ist
ic
s
,
2016,
pp
.
370
-
3
78.
[19
]
A.
Rai
,
B.
Aw
a
sthi,
S.
Singh,
a
nd
C.
K.
Dw
ive
di,
“
A
rev
ie
w
of
m
axi
m
u
m
po
wer
point
tra
ck
i
ng
te
chni
qu
es
for
photovol
taic
s
y
st
em,
”
In
te
rnation
al
Journal
of
En
gine
ering
Re
s
ea
rch
,
vol
.
5
,
no
.
6
,
pp
.
539
-
545
,
2
016.
[20
]
L.
Thi
aw
,
G.
Sow
,
and
S.
Fal
l,
"
Applic
a
ti
on
of
n
eur
al
net
works
t
ec
hniqu
e
in
ren
e
wable
ene
rg
y
s
ystems
,
"
in
2014
Fi
rs
t
Int
ernati
on
al
Conf
ere
nce o
n
Syste
ms
In
formatic
s,
Mod
ellin
g,
and
Simulatio
n
,
W
ashingt
on
DC,
US
A,
2014.
[21
]
N.
A.
Kam
arza
m
an
and
C.
W
.
Ta
n
,
“
A
comprehe
nsive
r
evi
ew
of
m
axi
m
um
power
point
tracki
ng
al
gor
it
hm
s
for
photovol
taic
s
y
st
ems
,
”
R
ene
wabl
e
and
Susta
inable
En
ergy
Revie
w
s
,
vol. 37, pp. 58
5
-
598,
2014
.
[22
]
V.
Lo
Brano,
G.
Ciul
la
,
and
M.
Di
Falc
o,
“
Artificial
neur
al
ne
tworks
to
pre
dic
t
t
he
power
output
of
a
pv
pane
l
,”
Inte
rnational
Jo
urnal
of Phot
oen
ergy
,
vo
l. 2014,
pp.
1
-
12
,
2014
.
[23
]
S.
Naz
m
ul
and
A.
Hojjat,
“
Sy
ner
g
ie
s
of
fu
z
z
y
logic,
neur
a
l
net
works
and
evol
ut
iona
r
y
c
om
puti
ng
,”
in
Computati
onal I
nte
lligen
ce
,
John
W
il
e
y
&
Sons
,
pp.
159
-
181
,
20
13
.
[24
]
V.
K.
Garg,
et
al.,
“
A
rev
ie
w
pap
er
on
var
ious
t
y
pes
of
MPPT
te
c
hnique
s
for
PV
s
y
stem,
”
Inte
rnat
ional
Journal
of
Engi
ne
ering
and
Scienc
e
R
ese
arc
h
(
IJE
SR)
,
vol
.
4
,
no
.
5
,
pp
.
320
-
330,
2014
.
[
2
5]
A.M.
Fara
y
o
la, A
.
N.
Has
an, a
n
d
A.
Al
i, “
E
ffici
ent
pho
tovol
t
ai
c
MPPT
sy
st
em using
co
arse
g
aussian
support
vec
tor
m
ac
hin
e and
ar
ti
fi
cial
n
eu
ral
n
et
work
te
ch
nique
s
,
”
Int
ernat
ional
Journal
of
Innov
ative
Com
puti
ng,
Information
and
Control
(
IJI
CIC)
,
vol
.
14
,
no
.
2
,
p
p.
323
-
339
,
201
8.
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
Op
ti
miz
atio
n o
f PV Sy
ste
ms U
sing D
ata Mi
ni
ng an
d
Re
gr
es
sion Le
ar
ne
r…
(
Adeday
o
M
. F
arayol
a
)
1089
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Adeda
y
o
M.
Fara
y
o
la
re
ceive
d
his
Bac
hel
or
of
Scie
nce
s
degr
e
e
in
El
e
ct
r
ic
a
l
and
El
e
ct
roni
c
Engi
ne
eri
ng
S
ci
en
ce
s
from
the
Univ
ersity
of
L
agos
i
n
2012,
MEn
g.
Degr
ee
in
El
e
ct
ri
ca
l
&E
l
ect
ronic
Eng
ine
e
ri
ng
Scie
nc
es
fro
m
the
Univer
si
t
y
of
Johanne
s
burg
in
2017
.
Curre
ntly
worki
ng
on
his
Phd
p
rogra
m
m
e
at
the
Univer
sit
y
o
f
Johanne
sburg,
South
Afric
a
.
His
rese
arc
h
int
er
est
s
inc
lude
Mac
h
ine
l
e
arn
ing
,
power
s
y
stems
o
pti
m
iz
ation
,
r
enew
abl
e
energ
y
s
y
stems
and
sm
a
rt
grid
.
Dr.
Ahm
ed
Ali.
He
rec
ei
v
ed
his
MS
c.
degr
ee
in
El
e
ct
ri
ca
l
and
E
l
ec
tron
ic
E
ngin
eering
from
the
Univer
sit
y
of
Jo
ha
nnesburg
and
his
PhD
degr
ee
i
n
El
e
ct
ri
cal
E
ng
i
nee
ring
from
the
Univer
sit
y
of
Johanne
sburg
in
2014
and
2017
r
espe
ctively
.
Cur
ren
tly
work
ing
a
s
a
l
ecture
r
with
the
U
niv
ersi
t
y
of
Johanne
sburg.
His
rese
arc
h
in
te
rests
are
th
e
power
sy
st
ems
oper
at
ion
and
pla
nn
i
ng,
el
e
ct
ri
ci
t
y
m
ark
et
an
aly
s
is,
distri
but
ed
generat
ion
,
power
s
ystem
opti
m
iz
atio
n,
Mac
hin
e
l
ea
r
ning
and
sm
art
grid
.
Evaluation Warning : The document was created with Spire.PDF for Python.