In
te
r
n
ation
a
l Jou
rn
al
o
f Po
we
r
Elec
tron
ic
s an
d
D
r
ive S
y
stem
(IJ
PED
S
)
V
o
l.
10, N
o.
3, S
ep 2019,
pp.
2
1
4
8
~2
1
5
6
ISSN: 2088-
8694,
DOI
:
10.11591
/ijpeds.
v10.
i
3.pp2148-2156
2
148
Jou
rn
a
l
h
o
me
pa
ge
:
ht
tp:
//i
a
e
score
.
com
/
j
o
u
r
na
l
s
/
i
n
d
e
x
.
p
hp/IJ
PED
S
Performance comparison of arti
f
i
cial inte
lligence technique
s
in
short term current forecastin
g
for photovoltaic system
Mu
h
a
mm
ad
M
u
r
tadh
a Oth
m
an
, Moh
a
mmad
Fazr
u
l
A
sh
raf
Moh
d
Faz
il,
Moh
d
H
af
ez Hilm
i Haru
n, Isma
i
l Mu
si
r
i
n,
Sha
hril
I
rwa
n
S
u
l
a
i
m
a
n
Fac
u
lty
o
f Ele
c
t
rica
l
En
gine
erin
g,
U
nivers
iti
Tek
n
o
l
ogi
M
A
R
A
,
M
alay
sia
Ar
ticle
In
fo
ABSTRACT
A
r
t
i
cl
e hi
sto
r
y:
Re
cei
v
e
d
No
v
2
7
, 2
018
Re
vise
d Ja
n 2
4
,
2019
Acc
e
p
t
e
d
M
a
r 1
3
, 2
019
T
h
is
p
a
p
er
p
res
e
nts
artifici
al
i
nte
lli
gen
ce
appro
ach
o
f
artific
ial
neu
r
al
n
etwo
rk
(A
NN)
a
nd
r
andom
f
o
r
est
(RF
)
t
hat
u
s
ed
t
o
perfo
r
m
s
hort
-
term
p
h
ot
ovo
lt
aic
(P
V
)
o
u
t
pu
t
cu
rrent
f
o
recasting
(S
TPCF
)
f
o
r
t
h
e
n
e
xt
2
4-h
o
u
r
s.
The
i
nput
da
ta
f
o
r
AN
N
and
R
F
i
s
consis
t
s
o
f
mul
tipl
e
time
l
a
gs
o
f
hour
ly
s
o
l
a
r
i
rradi
ance,
t
e
m
p
eratu
r
e,
hour,
p
o
w
e
r
an
d
cu
rrent
t
o
d
e
termin
e
th
e
m
ovem
e
n
t
pa
t
t
ern
o
f
d
a
ta
t
h
a
t
hav
e
b
een
d
en
oi
sed
b
y
u
s
i
n
g
w
av
elet
d
ecom
p
o
s
i
t
i
on.
T
h
e
L
e
venb
erg-
M
a
rq
uardt
optimizati
on
tech
niq
u
e
is
u
sed
as
a
b
ack
-pro
pagat
i
o
n
al
g
o
ri
t
h
m
f
o
r
A
NN
and
the
b
a
g
g
i
n
g
b
a
s
e
d
boot
st
rappi
ng
t
ech
n
i
qu
e
is
u
s
e
d
in
t
h
e
RF
t
o
i
m
prov
e
th
e
resu
lts
o
f
f
o
recas
ting.
T
h
e
i
nf
orm
a
ti
on
o
f
P
V
outpu
t
current
i
s
o
b
t
a
ined
f
rom
Green
E
nergy
Res
earch
(
G
E
RC)
U
ni
vers
it
y
Techn
o
l
o
g
y
Mara
S
h
ah A
lam, M
alaysi
a and
is
u
sed
as t
he cas
e st
udy
i
n es
tim
atio
n
o
f
PV
o
ut
put
cu
rrent
f
o
r
t
he
n
e
x
t
2
4
-ho
u
rs.
Th
e
res
u
lts
h
av
e
sh
ow
n
th
at
both
p
ro
po
se
d
t
echn
i
qu
es
a
re
a
bl
e
t
o
p
erf
o
rm
f
o
r
ecast
i
ng
o
f
f
u
ture
hou
rly
P
V
o
u
tput
c
urrent
w
i
t
h
l
ess error.
Key
w
ords:
Ar
t
i
ficia
l
n
eural
netw
ork
Mu
l
tip
le
tim
e
l
a
gs
Ra
nd
om
for
est
S
hort
ter
m
p
h
o
to
vo
l
t
ai
c
current
for
eca
s
t
i
n
g
Wave
le
t de
no
isi
n
g
Co
pyr
i
g
h
t
©
2019
Instit
u
t
e of Advan
ced E
n
g
i
neer
in
g a
nd Sci
e
nce.
All rights
reserv
e
d
.
C
o
rre
s
po
n
d
in
g Au
t
h
or:
Mu
ham
m
a
d
M
urta
dha
O
t
h
m
a
n,
F
acult
y o
f
E
lec
t
ric
a
l
En
g
i
n
eer
ing,
U
n
i
v
er
si
ti Te
k
n
o
l
og
i MA
RA
,
40
4
50 S
h
ah A
la
m, S
elang
o
r,
Ma
l
a
y
sia.
Em
ail
:
m
am
at505my
@
ya
ho
o.
com
1.
I
N
TR
OD
U
C
TI
O
N
The
pa
s
t
f
ew
y
ear
s
hav
e
s
h
o
w
n
a
r
em
arka
bl
e
grow
t
h
i
n
t
h
e
use
of
s
olar
e
ne
r
gy
fo
r
residen
tia
l,
com
m
e
r
cia
l
,
a
nd
i
ndus
tria
l
sectors.
T
he
g
ro
w
i
n
g
c
apac
i
t
y
for
gl
o
b
al
s
o
l
a
r
P
V
sec
t
or
a
lrea
dy
rea
c
he
d
1
78G
W
in
2
0
1
4
,
and
e
s
t
i
m
a
te
d
to
r
e
a
c
h
5
40G
W
i
n
2
0
1
9
[
1,
2
].
I
n
rec
e
nt
y
ea
rs,
solar
P
V
s
yste
m
has
bee
n
d
e
v
el
o
p
ed
drast
i
ca
ll
y an
d
the re
aso
n
b
e
h
ind it is bec
a
u
s
e
of t
h
e
na
t
u
re
u
sa
g
e
o
f
P
V
t
hat is m
ain
t
e
n
a
n
ce
free
, l
on
g la
st
i
n
g
use
d
,
a
nd
en
vir
onm
en
ta
l
l
y
frie
nd
l
y
[
3-
8]
.
H
o
w
e
ver,
P
V
syst
em
i
s
opera
tin
g
in
a
n
o
n
-sta
t
i
o
n
ary
r
a
nd
om
p
roce
ss
ca
use
d
b
y
t
h
e
v
a
ria
b
i
l
i
t
y
o
f
s
o
l
a
r
irr
a
d
i
an
ce
a
n
d
o
t
h
e
r
e
n
v
i
r
o
n
m
e
n
t
a
l
f
ac
to
rs
t
ha
t
ma
y
affec
t
t
he
o
u
t
p
u
t
c
urr
e
nt.
In
g
e
n
era
l
,
a
r
tific
ia
l
ne
ural
n
e
t
w
o
rk
(
A
N
N
),
s
upp
ort
ve
c
t
o
r
m
ac
hine
(
SVM),
a
nd
fuz
z
y
log
i
c
hav
e
bee
n
u
se
d
a
s
f
o
r
ec
astin
g
m
e
tho
d
s
due
t
o
sever
a
l
adva
n
t
ag
es
[
9,
10].
These
me
t
hods
h
a
v
e
be
en
u
se
d
for
solar
irra
d
i
a
n
ce
fore
ca
st
in
g
d
u
e
t
o
its
i
nc
rea
s
i
n
g
dem
a
nd
in
p
rod
u
c
i
n
g
acc
ur
ate
o
u
t
put.
O
t
her
tha
n
a
n
y
b
asic
A
I
tech
n
i
q
u
e,
t
her
e
i
s also
a
c
ombi
na
ti
o
n
of sev
e
r
a
l
A
I
t
ec
hn
i
q
ue
s
t
ha
t can prod
u
ce
a
ccur
a
te
forec
asti
ng re
sul
t
i
n
the
fu
t
u
re
o
f
solar
irra
di
a
n
ce.
A
tim
e
ser
i
e
s
w
it
h
A
N
N
,
fuzz
y
l
o
g
i
c
w
it
h
A
N
N
a
nd
w
a
v
e
l
e
t
base
d
A
N
N
a
r
e
exa
m
ple
s
o
f fa
m
o
us
c
om
b
i
na
tio
n
for A
N
N
.
I
n
p
a
r
t
i
c
u
l
a
r
,
A
N
N
i
s
a
n
a
l
t
e
r
n
a
t
i
v
e
m
o
d
e
l
t
h
a
t
c
a
p
a
b
l
e
o
f
h
a
n
d
lin
g
u
n
ce
rta
i
n
t
y
m
a
tter
s
o
f
solar
i
rrad
i
a
n
ce
[
1
1
,
1
2
]
.
Th
e
mai
n
a
d
v
a
nt
ag
es
o
f
u
s
i
n
g
ANN
can
b
e
s
e
en
i
n
it
s
st
a
b
il
it
y
t
o
s
olv
e
c
om
ple
x
m
od
elli
ng
espec
i
al
ly
a
non-l
i
nea
r
m
odel
[9].
R
and
o
m
fore
st
(
RF
)
is
a
not
he
r
adva
nc
e
A
I
u
sed
for
forec
a
s
tin
g.
R
F
uses
ensem
b
le
m
achi
n
e
le
ar
ni
n
g
t
ha
t
c
o
n
s
is
ts
o
f
ma
n
y
d
e
c
i
s
i
o
n
tre
e
m
ode
ls
f
or
c
lassifica
t
i
o
n
and
regr
essio
n
[
13]
.
The
c
o
nstruc
ti
on
o
f
t
r
ee
doe
s
not
d
e
p
e
nd
o
n
t
he
p
re
vio
u
s
tre
e
s
i
n
c
e
t
h
e
t
r
e
e
s
a
r
e
c
r
e
a
t
e
d
i
n
d
e
p
e
n
d
e
n
t
l
y
b
y
u
s
i
n
g
bo
o
t
s
t
ra
p
a
ggr
ega
t
i
on
tec
h
ni
que
a
s
w
e
l
l
a
s
the
bag
g
i
n
g
[14,
15]
.
RF
h
as
t
he
a
dva
n
t
a
g
e
i
n
t
er
ms
o
f
non-
Evaluation Warning : The document was created with Spire.PDF for Python.
Int J
P
o
w
E
l
e
c
&
D
ri S
yst
IS
S
N
:
2088-
86
94
Perf
om
a
n
ce
co
m
par
is
on o
f
ar
tif
ic
ia
l i
n
te
l
l
i
g
e
n
c
e
te
ch
n
i
ques
in sh
ort te
rm
… (
M
uh
am
m
a
d Mur
t
a
d
ha O
t
hm
a
n
)
2
149
overf
itt
i
ng
t
he
o
u
t
pu
t
r
e
sul
t
s,
t
he
r
un
tim
e
pro
c
ess
is
f
ast
a
nd
e
f
fic
i
e
n
t
w
h
en
h
an
d
l
i
n
g
a
lar
g
e
data
se
t
th
us
g
ive
s
it
s
uperi
or
p
red
i
ct
i
v
e pe
rfor
m
ance.
This
p
aper
p
r
e
sents
the
ANN
an
d
RF
m
et
hods
t
h
a
t
used
t
o
perform
s
ho
r
t
-
t
e
r
m
P
V
out
pu
t
c
u
rre
nt
fore
ca
sti
n
g
for
the
nex
t
2
4-h
ours.
T
here
i
s
no
t
muc
h
r
esea
rch
t
ha
t
has
b
e
en
d
o
n
e
re
gardi
n
g
to
t
he
P
V
ou
tp
ut
curr
ent
for
eca
s
t
i
n
g
us
i
ng
RF
.
The
in
pu
t
da
t
a
u
se
d
for
th
is
m
eth
o
d
i
s
c
u
r
r
ent,
i
rradia
n
ce,
hours
a
n
d
tem
p
e
r
a
t
u
r
e
w
ill
pass
thro
u
gh
the
filtra
t
i
o
n
proce
s
s
usi
n
g
t
h
e
w
a
vele
t
dec
o
mp
os
i
tio
n
to
e
limi
n
ate
t
h
e
no
ises
i
n
e
a
c
h
da
t
a
.
The
n
,
the
m
u
l
t
ip
le
t
im
e
lags
i
s
used
due
t
o
it
s
ca
pabi
l
i
t
i
es
t
o
i
de
nt
i
f
y
t
h
e
pa
tte
rn
a
n
d
b
e
h
av
i
o
r
of
f
ilter
e
d
dat
a
w
h
i
l
e
im
pro
v
i
ng
i
t
f
or
acc
urate
e
s
t
i
m
a
t
i
o
n
i
n
the
ne
x
t
2
4
ho
urs
o
f
P
V
out
p
u
t
c
u
r
r
ent
forec
a
st
in
g
[16-
20
]
.
T
he
ca
se
s
tu
dy
u
se
s
PV
out
p
u
t
c
ur
rent,
t
e
mpe
r
at
ure
,
i
rradia
n
c
e
a
n
d
hour
s
i
n
201
5
w
i
t
h
t
he
t
o
t
a
l
o
f
74
6
0
h
o
u
rl
y
data
o
b
t
a
i
n
ed
f
ro
m
t
h
e
Gree
n
En
e
r
g
y
R
e
sea
r
ch
C
en
t
e
r
(GERC
)
,
Un
iv
e
r
s
i
t
y
T
e
c
h
n
o
l
o
g
y
M
a
r
a
S
h
a
h
A
l
a
m
,
M
a
l
a
y
s
i
a
.
The
r
o
b
u
stne
ss
o
f
b
o
th
m
ode
ls
i
n
f
o
rec
a
st
i
n
g
a
r
e
com
p
are
d
by
re
fer
r
in
g
t
o
t
he
m
ea
n
sq
u
a
re
e
rr
or
(
MS
E),
me
an
abs
o
l
u
t
e
pe
r
ce
n
t
a
g
e e
rror (MA
P
E) a
nd re
gressi
on be
tw
e
e
n
the
for
eca
ste
d
a
nd
act
ua
l
(targe
t
e
d) va
l
ue
s.
2.
RESEARCH
M
ETH
O
D
Th
is
s
e
g
me
nt
w
il
l
e
x
p
l
a
i
n
th
e
conc
ep
t
of
f
ea
t
u
r
e
e
xtra
ct
io
n
or
data
p
re
para
t
i
on
for
the
A
N
N
a
nd
RF
mode
l
s
u
se
d
i
n
t
he
P
V
ou
tpu
t
c
ur
rent
f
or
eca
sti
ng
[2
1-2
5
].
T
he
s
truc
ture
u
se
d
for
th
is
p
r
o
ce
ss
i
s
sh
ow
n
in
F
igure
1
w
h
er
ei
n
the
S
T
P
C
F
proce
s
s
be
g
i
ns
f
r
o
m
t
h
e
or
i
g
i
n
al
d
a
t
a
se
lec
ti
o
n
.
In
t
h
i
s
case
,
t
he
h
our
ly
P
V
ou
t
p
u
t
c
urr
e
nt,
tem
p
era
t
ur
e,
i
rra
d
ia
nce
a
nd
hour
s
a
r
e
selec
t
ed
a
s
th
e
i
n
pu
t
dat
a.
A
fter
d
at
a
se
l
e
cti
o
n,
t
he
d
ata
prepa
r
ati
on
for
the
in
put
a
n
d
t
ar
get
da
ta
i
s
p
e
rfor
me
d
by
u
s
i
ng
the
w
a
ve
l
e
t
dec
ompos
i
t
i
on
an
d
m
u
l
tip
l
e
time
la
gs
t
e
c
h
n
i
que
.
S
ubseq
ue
n
t
l
y
,
t
h
e
fore
cas
ti
n
g
m
ode
ls
f
or
A
NN
a
nd
RF
a
re
d
esi
gne
d.
F
inal
l
y
,
the
tra
i
n
i
n
g
a
nd
te
st
i
ng
pro
c
e
dur
e
is
p
erf
o
rm
ed to
o
b
ta
in
t
he
for
e
c
a
s
tin
g o
u
tc
om
e
from
A
N
N
and
RF
.
F
i
gure
1. Bl
o
c
k
d
ia
gram
of PV
out
pu
t
curr
ent for
eca
sti
ng
m
odel
i
ng
2
.
1
.
I
nput da
t
a
o
f
chro
n
o
l
ogi
cal
pa
ra
mete
r
The
in
form
at
i
on
of
d
a
t
a
use
d
i
n
fore
ca
sti
n
g
i
s
a
cqu
i
red
fr
om
t
h
e
Gr
een
E
ne
rgy
R
e
se
ar
ch
C
e
n
tre
(G
ERC)
o
f
U
i
TM
S
ha
h
A
l
am
,
Mala
ysia.
The
data
i
s
ob
tai
n
ed
i
n
the
f
or
m
of
M
A
TLA
B
softw
a
r
e
.
The
da
t
a
i
s
ob
ta
ine
d
c
o
n
s
i
stin
g
w
i
t
h
f
ive
pa
ram
e
ter
s
i
n
201
5.
I
n
the
G
E
RC
l
a
bora
t
or
y,
t
h
i
s
in
for
m
a
tio
n
is
c
o
l
lec
t
e
d
b
y
da
ta
lo
gge
r
for
eve
r
y
5
m
i
nute
s
.
T
h
i
s
d
ata
w
ill
be
a
nal
y
ze
d
and
the
ho
ur,
irra
di
a
n
ce
,
tem
p
er
atu
r
e,
pow
er
a
nd
current
with
m
a
x
imum
pa
r
am
eter
value
wil
l
be
u
se
d
in forec
as
t
i
n
g
a
s sho
w
n
i
n
Ta
ble
1
.
Ta
b
l
e
1.
Informa
ti
o
n
f
or
eac
h pa
ram
e
ter
P
a
r
a
me
t
e
r
Ma
xi
m
u
m Va
lu
e
Unit
H
our
2
4
H
our
Irra
d
i
a
n
c
e
1320
W
/
m
²
Tem
p
e
r
a
t
ure
47.
9
°C
Pow
e
r
2684
7.
61
W
a
t
t
C
u
rre
nt
9
.98
A
m
p
e
re
The
da
ta
pre
pa
ra
t
i
on f
o
r A
N
N
and
RF
m
od
els is s
h
o
w
n
i
n
F
i
gure
2
an
d
i
t
s
pr
o
ced
ur
e is
e
lab
o
rate
d
as
fo
l
l
ow
s.
F
i
gur
e 2.
Bl
o
c
k
dia
gr
am
for
data
pre
p
a
r
at
io
n
Ra
w
d
a
t
a
o
f
hou
r
s
,
tempera
t
ur
e
,
irra
dia
n
ce,
cur
r
e
n
t a
n
d
po
w
e
r
Wa
vel
e
t
decom
p
os
itio
n
Dat
a
n
o
rm
a
lized
Mu
ltiple
Ti
m
e
L
ags
Dat
a
Ar
ra
ng
e
m
ent
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN: 2088-8694
Int J
P
o
w
El
e
c
&
D
ri S
yst
,
V
ol.
10,
N
o.
3
, S
e
p
2
0
1
9
:
214
8
– 2
156
2
150
-
Colle
ct
t
he
r
aw
d
ata
c
onsis
ti
n
g
w
it
h
7
460
hou
r
l
y
i
n
f
o
rmatio
n
of
h
o
u
r,
i
rra
di
a
n
ce,
t
emper
a
t
u
re,
pow
er
a
nd
curr
ent.
- P
e
r
f
orm the
fil
t
e
r
i
n
g proc
ess
of
i
n
put
d
a
t
a b
y
u
s
i
n
g
t
he
w
a
v
e
le
t de
c
o
m
p
o
s
i
tio
n
to
r
e
duc
e
any no
ise
ins
i
de
it.
-
N
o
r
m
a
l
iz
e
all
of
t
he
f
i
l
t
e
r
ed
d
a
t
a
by
d
i
v
i
din
g
w
ith
i
ts
m
a
x
im
um
v
alu
e
i
n
orde
r
to
r
e
duce
d
a
ta
r
e
dun
da
ncy
w
ithi
n
t
he
ra
n
g
e
of 0 a
nd 1.
-
P
e
rform
t
he
m
ult
i
p
le
t
im
e
lags
t
o
impr
o
v
e
the
data
t
o
de
te
rm
i
ne
t
he
m
ov
e
m
e
n
t
pa
t
t
e
r
n
of
e
very
i
np
ut
d
ata
requ
ire
d
by
the
A
N
N
and RF
t
ec
hni
que
s.
-
U
s
e
the
mult
i
p
le
t
ime
la
gs
t
o
e
s
tim
ate
t
h
e
fu
ture
v
aria
bl
e
a
n
d
t
h
e
lag
g
e
d
(pa
s
t
peri
o
d
)
varia
b
l
e
t
ha
t
w
ill
evo
l
ved
i
n
t
he
f
u
t
ure
[9,
9,
9
,
9,
9
].
T
he
i
n
p
u
t
d
a
t
a
impr
ove
d
b
y
t
he
m
ul
t
i
ple
t
i
me
l
a
g
sa
b
l
e
to
d
e
t
er
mine
t
he
move
me
nt
p
a
tte
rn
o
f
da
t
a
i
n
t
h
e
ne
ural
n
e
t
w
o
r
k
b
a
s
e
d
o
n
(1).
T
h
e
to
tal
n
u
mbe
r
o
f
t
i
me
i
nte
r
va
l
la
g
g
i
n
g
i
s
K
=
24
h
ours.
L
agk
=
Zt –
Zt
-k
(1
)
wher
e,
t
: time
i
n
terva
l
.
k
:
t
i
m
e
i
n
t
e
r
v
al
l
ags
1
,
2
,
3
…
K
.
K
:
t
o
ta
l
num
b
e
r
of
t
i
m
e
in
t
e
rval
l
a
g
gin
g
.
I
n
(
1),
the
to
ta
l
numbe
r
of
t
i
m
e
inter
v
a
l
l
a
g
g
i
ng
use
d
i
s
K
=
2
4
w
h
e
r
e
the
val
u
e
of
K
i
s
st
a
t
e
d
t
o
be
equ
i
vale
n
t
to
the
time
in
terva
l
i
n
the f
o
rec
a
sted va
r
ia
b
l
e.
The
v
al
ue
o
f K
i
s
fixe
d
t
o 24 f
o
r forec
a
sti
n
g
th
e
next
24
ho
urs
of
P
V
ou
tpu
t
c
urr
e
nt
.
The
inpu
t
da
t
a
i
s
i
n
t
he
f
orm
of
k
-b
y-t
m
a
trix
w
here
e
ac
h
col
u
m
n
w
i
ll
be
u
se
d
to
f
or
eca
s
t
P
V
ou
t
put curr
e
n
t
f
or
t
he ne
x
t 2
4
-
h
o
u
rs. The first
co
lum
n
o
f train
i
ng da
t
a
,
L
agk
is
us
ed to f
o
r
e
ca
st
the ta
r
g
et da
t
a
of
X48
.
The
i
n
pu
t da
t
a
a
rrang
e
m
e
n
t
for
trai
n
i
n
g
an
d
t
a
r
ge
t
data i
s sh
ow
n
in
F
igure
3.
F
i
gur
e 3.
The
inp
u
t
d
ata
arr
a
n
g
em
ent
for tra
i
nin
g
d
ata a
n
d t
a
r
g
e
t
da
t
a
for
A
N
N
a
nd RF
in La
g 2
4
T
w
o
s
e
t
s
o
f
d
a
t
a
w
h
i
c
h
i
s
t
r
a
i
n
i
n
g
d
a
t
a
a
n
d
t
a
r
g
e
t
d
a
t
a
i
s
c
r
e
a
ted
afte
r
all
data
h
a
v
e
bee
n
c
onver
t
e
d
i
n
t
o
m
u
l
t
i
p
l
e
t
i
m
e
l
a
g
s
.
T
h
e
a
r
r
a
n
g
e
m
e
n
t
o
f
t
r
a
i
n
i
n
g
d
a
t
a
f
o
r
e
ac
h
line
is
a
c
ombina
ti
o
n
o
f
ho
ur,
t
e
mpe
r
ature
,
irra
d
i
a
n
ce
an
d
cur
r
ent.
T
he
l
a
s
t
par
t
o
f
c
u
rr
ent
da
ta
w
ill
be
u
se
d
as
t
a
r
ge
t
data
.
The
t
r
ai
nin
g
a
n
d
t
ar
ge
t
da
ta
form
ed
w
ill
be
u
sed
for
forecast
i
n
g
using
ANN
and
R
F
m
ethods.
T
he
c
hro
n
o
lo
gica
l
ar
rang
em
ent
of
i
np
u
t
d
a
t
a
is
s
h
o
w
n
i
n Ta
bl
e
2.
Tab
l
e
2.
C
hro
n
o
l
o
gica
l
arra
ng
em
ent of i
n
put
d
ata
In
p
u
t
d
a
ta
Par
a
m
e
t
e
r
f
o
r
e
ach
l
i
n
e
af
t
e
r
L
a
g
2
4
1
to
2
3
hou
r
24
to
4
6
tem
p
e
r
a
t
ur
e
47
to
6
9
irr
a
dia
n
c
e
70
to
9
2
c
u
rre
nt
Ta
r
g
et
D
a
t
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t
J
P
o
w
Elec
&
D
r
i
S
y
st
I
S
S
N
:
2088-
86
94
Pe
rf
om
a
n
c
e
c
o
m
paris
on o
f
arti
f
i
c
i
a
l
inte
l
l
i
g
e
n
ce
tec
h
ni
que
s
in sh
ort
te
rm
… (
M
uham
m
a
d Mur
t
a
d
ha O
t
hm
a
n
)
2
151
C
u
rre
nt
2
.
2
.
Art
i
ficial
n
eura
l
netw
o
r
k
(
ANN)
The
Arti
fic
i
a
l
N
eura
l
Network
(ANN)
is
a
n
alter
n
a
t
i
v
e
m
e
tho
d
t
h
a
t
h
as
b
e
e
n
eff
i
c
i
en
t
l
y
car
ried
o
u
t
i
n
th
is
p
a
p
er
a
s
it
is
a
ls
o
su
i
t
ed
t
o
tac
k
l
e
s
ola
r
e
nergy
unce
r
tai
nty
issu
es.
Th
e
ANN
is
u
s
i
n
g
th
e
Lev
e
n
b
e
r
g
-
Mar
q
uar
d
t
te
ch
ni
que
t
o
f
o
r
e
c
a
s
t
t
h
e
P
V
outp
u
t
c
u
r
r
e
n
t
f
or
t
h
e
n
ext
24
h
ours [4
].
I
n
t
his
case
s
tu
dy
,
t
h
e
ANN
mo
d
e
l
fo
r
fo
r
ecasting
th
e
P
V
o
utput
c
ur
r
e
n
t
i
s
co
nsist
i
ng
of
one
i
np
ut
laye
r
,
t
w
o
h
idd
e
n
la
yer
s
,
and
one
o
u
t
pu
t
la
ye
r
.
T
he
out
pu
t
la
ye
r
o
f
A
N
N
i
s
c
o
n
s
i
s
t
i
n
g
o
f
o
n
e
n
e
u
r
o
n
w
h
i
c
h
w
i
l
l
pr
ovi
des
t
h
e
pr
e
d
i
c
t
e
d
P
V
out
put
c
ur
r
e
nt
f
or
t
he
n
e
x
t
2
4
h
o
u
r
s
.
T
h
e
L
e
v
e
n
b
e
r
g
-
M
a
r
q
u
a
r
d
t
t
e
c
h
n
i
q
u
e
i
s
u
s
e
d
i
n
the
A
N
N
mode
l
a
s
b
a
c
k
pr
o
p
aga
tio
n
al
g
o
r
i
thm
f
o
r
o
p
tim
i
zati
o
n
of
t
h
e
d
ata
d
u
rin
g
t
he
t
rai
n
i
n
g
p
r
oces
s.
T
his
tec
h
n
i
q
u
e
is
c
om
monl
y
use
d
i
n
for
eca
s
t
i
n
g
t
h
e
t
r
ai
nin
g
s
et
o
f
A
N
N
due
t
o
it
s
a
l
g
o
r
i
thm
t
h
a
t
c
ompr
om
ise
be
t
w
ee
n the
a
ccur
acy an
d
st
a
bil
i
t
y
o
f pred
ic
ti
o
n
t
o a
c
h
i
eve
the
st
eep
es
t
method
f
o
r
measu
rin
g
min
i
mal
err
o
rs.
The
A
N
N
m
odel
for
for
e
ca
st
ing
P
V
outp
u
t
cur
r
e
nt
i
s
show
n
in
F
ig
ur
e
4
and
the
AN
N
pr
oce
dur
e
is
e
x
p
l
a
i
ne
d
be
l
o
w
.
-
D
i
v
i
de
i
n
p
u
t
d
a
t
a
i
n
to
t
hr
ee
s
ets
o
f
t
r
a
in
i
n
g
,
t
e
s
t
i
ng
an
d
va
l
i
da
ti
n
g
f
or
t
he
m
ulti
ple
t
i
me
l
ags
of
K
=
2
4
h
our
s.
-
I
n
t
he
t
r
a
i
n
i
n
g
pr
ocess,
t
he
s
yna
pses
m
in
i
m
ize
the
er
r
o
r
be
t
w
ee
n
t
h
e ac
tua
l
o
ut
put
a
n
d
t
h
e t
a
rg
e
t
ed
o
utp
u
t
b
y
r
e
gu
lat
i
ng
t
h
e
l
e
a
r
ni
ng
r
a
te
a
n
d
m
ome
n
t
u
m
.
-
S
e
lec
t
t
he
n
um
ber
of
h
id
de
n
l
a
yer
s
i
s
bas
e
d
on
t
h
e
fa
c
t
o
f
o
n
e
h
i
dd
en
l
ay
e
r
i
s
suf
f
i
c
i
e
n
t
t
o
est
i
m
a
t
e
a
n
y
f
u
n
c
tio
n.
T
her
e
for
e
,
tw
o
hid
d
e
n
l
aye
r
s
is
u
se
d
in
t
his
A
N
N
m
ode
l
s
t
ha
t
w
i
l
l
p
r
o
vi
de
m
or
e
pr
ecise
r
esults
w
i
t
h
m
i
ni
m
u
m
R
M
S
er
r
o
r
in
f
or
ec
a
s
tin
g
the
nex
t
2
4
h
o
u
r
s
o
f
P
V
out
p
u
t
c
u
r
re
n
t
.
-
R
e
p
ea
t
th
e
er
ro
r
mi
n
i
mi
z
a
tio
n
p
r
o
c
ess
u
n
t
i
l
t
h
e
opt
i
m
i
zati
o
n
p
r
o
cess
i
n
f
o
r
ec
a
s
ti
ng
i
s
co
nv
e
r
g
e
d
yi
e
l
di
ng
t
o
the
sm
al
l
e
s
t
e
r
r
or
i
n
i
t
s
o
u
t
p
u
t
.
T
he
n,
t
he
t
r
a
ini
n
g
pr
oce
dur
e
i
s
t
erm
i
na
te
d once t
h
e
min
i
mum
err
o
r
be
com
e
s
pla
t
e
a
u
f
or
s
ev
e
r
al
i
te
r
a
t
i
ons
o
f
op
tim
i
z
a
t
io
n
pr
oce
s
s
in
v
o
lv
e
d
in
th
e
ANN.
-
I
d
en
tify
t
h
e
s
t
r
eng
t
h
of
ANN
in
p
r
o
d
u
c
ing
th
e
co
rrect
S
TP
C
F
r
esults
t
hat
c
a
n
b
e
pr
o
v
e
n
by
c
ond
uc
t
i
n
g
t
he
tes
t
i
n
g
the
n
v
a
l
i
d
a
tio
n
pr
oce
s
s
e
s
by
us
in
g
d
i
f
f
e
r
ent
se
t
of
i
n
p
ut d
a
t
a.
F
i
gur
e
4.
B
l
o
c
k
d
i
a
gr
am
f
or
A
N
N
m
odel
2.
3.
R
a
n
d
om
f
ore
s
t
(
R
F)
Ra
nd
om
F
or
e
s
t
is
a
m
ode
l
c
o
mpr
i
si
ng
w
ith
t
w
o
s
i
g
n
i
fica
n
t
c
om
pon
e
n
ts
o
f
tr
ee
bagg
i
ng
and
r
a
nd
om
de
ci
sio
n
t
r
e
es
[
6]
.
The
Tr
e
eBagger
def
i
ne
d
as
B
is
c
on
t
a
in
in
g
w
i
t
h
t
he
n
um
ber
of
t
r
e
es
(
NTre
es
)
with
t
he
X
a
n
d
Y
a
s
the
e
n
sem
b
le
f
u
n
c
tio
n
th
a
t
b
ee
n
use
d
f
o
r
c
re
ating
a
d
e
c
i
s
i
o
n
t
r
ee.
T
he
d
eci
si
o
n
t
r
e
e
u
s
es
t
he
i
n
p
u
t
fu
nc
ti
on
X
to
p
r
e
di
c
t
t
h
e
t
a
r
get
r
e
spo
n
s
e
Y.
T
he
p
r
o
c
e
dur
e
of
R
a
n
d
o
m
F
o
r
e
s
t
is
e
xpla
i
ned
be
l
o
w
.
-
P
e
r
f
or
m
boo
t
s
tr
ap
s
a
m
ples,
N
r
a
ndom
ly
d
r
a
w
n
f
r
o
m
the
tr
a
i
n
i
n
g
d
a
t
a
of
R
F
mode
l
,
t
o
cr
eate
a
r
e
gr
e
ssi
o
n
tre
e
s for e
ach
s
am
ple.
T
he
b
o
o
t
s
trap
s
am
p
l
e
is ha
v
in
g
t
h
e
s
a
m
e
si
z
e
a
s the
orig
ina
l
tra
in
i
ng da
t
a
.
-
P
e
r
f
or
m
t
h
e
bag
g
i
n
g
t
e
c
h
n
i
que
t
ha
t
di
vi
de
s
t
h
e
bo
otstr
a
p
sa
mp
le
i
nt
o
tw
o
sets
o
f
data
w
hic
h
i
s
tw
o-
t
h
i
r
d
is
fo
r th
e
In
-Bag
w
h
ile th
e
r
emaini
ng
da
ta
i
s
f
o
r
the
O
u
t
-
O
f
-
B
ag
(
OOB)
.
-
Use
the
InBa
g
t
o
c
re
ate
a
fo
re
st
w
he
rein
t
h
e
t
r
ee
growth
t
ech
n
i
q
u
e
w
i
l
l
p
r
o
d
u
c
e
t
h
e
b
e
s
t
l
e
a
v
e
s
.
.
T
h
e
O
O
B
da
t
a
i
s
use
d
t
o
r
un
the
u
n
b
i
a
s
e
d
p
r
e
dic
t
io
n
er
r
o
r
as
t
r
e
e
s
a
re
a
dded
i
n
to
t
h
e
f
or
est
dur
in
g
tr
ee
gr
ow
th
p
has
e
us
in
g
the
I
n
Ba
g
data
.
The
pr
im
ar
y
r
o
le
o
f
O
O
B
da
t
a
i
n
t
r
ee
gr
ow
t
h
t
ech
ni
qu
e
i
s
t
o
co
mp
are
i
t
s
e
sti
m
at
ion
w
i
t
h
t
he
p
r
e
d
i
cted
v
a
l
ue
s
o
b
t
ai
ne
d
fr
om
t
h
e
I
nBag
t
o
f
i
n
d
t
h
e
b
e
s
t
lea
v
es
w
ith
m
in
im
al
e
r
r
or
r
a
t
e
from
e
v
ery
t
r
ee.
-
H
a
l
t
t
he
g
r
o
w
t
h o
f
the
t
r
e
e once the f
i
nal
n
ode
o
f bes
t
l
e
a
f
in
eve
r
y
tr
ee
is
o
bta
i
ned.
U
pon f
i
ni
s
h
in
g
t
h
e fin
a
l
no
de
s,
t
he
p
r
e
di
c
tio
n
val
u
e
f
r
o
m
the
f
i
na
l
no
de
o
f
best
l
ea
f
i
s
c
o
l
l
e
ct
ed
f
ro
m
e
v
ery
tree
a
n
d
t
h
e
ave
r
ag
e
pr
ed
ict
i
on
i
s
c
a
l
c
u
l
a
te
d
f
r
o
m
the
f
i
n
a
l
no
de
l
e
a
f
of
a
l
l
t
r
e
e
s
.
F
igur
e
5
show
s
the
st
r
u
c
t
ur
e
of
R
F
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN: 2088-8694
Int J
P
o
w
El
e
c
&
D
ri S
yst
,
V
ol.
10,
N
o.
3
, S
e
p
2
0
1
9
:
214
8
– 2
156
2
152
F
i
gure
5. S
tructure
of
RF
a
l
g
o
r
it
hm
3.
RESULT
S
A
N
D
ANALY
S
IS
Th
is se
c
tio
n
disc
usse
d on the
S
T
P
C
F resul
t
s
deter
m
i
n
ed
b
y us
i
n
g
t
he A
N
N
a
n
d
R
F
m
odel
s
.
The
data
of
h
our
ly
s
ola
r
i
rradia
nce,
t
e
m
pe
rature
,
hour,
and
curr
ent
ob
t
a
i
n
e
d
f
ro
m
Gree
n
Energy
Re
sea
r
ch
(
GERC)
Uni
v
ersi
ty
T
ec
hn
o
l
o
g
y
M
ara
Shah
A
la
m,
M
a
l
ays
i
a
is
u
se
d
for
the
case
stud
y
of
S
TP
CF
.
The
inp
u
t
da
ta
un
derg
oe
s
t
h
e
w
a
ve
l
e
t
dec
o
m
pos
i
tio
n
to
e
li
m
i
na
t
e
t
he
n
oi
se
i
ns
i
de
t
h
e
d
a
t
a
and
t
h
e
n
t
h
e
m
ulti
ple
time
la
gs
o
f
K
=
24-
h
o
u
rs
i
s
a
p
p
lie
d
t
o
t
he
f
il
ter
e
d
da
ta.
The
da
ta
s
ize
u
s
ed
i
n
A
N
N
a
n
d
R
F
p
r
o
c
e
d
u
r
e
i
s
1
7
5
2
0
c
o
l
u
m
n
s
.
The
da
ta
i
s
d
i
vide
d
i
n
t
o
t
hr
ee
s
e
ts
w
her
e
i
n
t
he
d
a
t
a
s
i
z
e
f
or
t
r
a
in
ing
is
5
7
85
c
o
lum
n
s,
t
es
t
i
n
g
data
i
s
7
2
0
c
o
l
umns
and
va
l
i
da
ti
on
da
ta i
s 72
0
co
l
u
mns.
3
.
1
.
Arti
f
i
c
i
a
l
n
e
u
r
a
l
n
e
t
wo
r
k
(
ANN)
The
input
data
o
f
ANN
i
s
t
he
c
om
bina
tion
of
m
ultiple
time
lags
of
hour
,
t
e
mpe
r
ature
,
i
rradi
a
n
ce
a
n
d
curr
ent. Train
i
ng a
n
d tes
t
i
n
g proce
dures of A
N
N a
r
e
perfo
rme
d
w
he
re
th
e
inp
u
t
da
t
a
ha
vi
n
g
t
he
m
u
l
t
i
p
le tim
e
lag
s
o
f
K
=
24
hours.
I
n
t
h
e
A
N
N
m
odel,
t
he
number
of
n
eur
ons
f
or
t
h
e
f
irst
h
id
de
n
laye
r
is
20
a
nd
seco
n
d
hi
dde
n
la
ye
r
is
10.
W
h
ile,
l
e
ar
n
i
n
g
r
ate
a
nd
m
o
me
ntum
i
s
0.
3
re
specti
v
e
l
y.
T
he
num
bers
f
or
f
i
r
st
h
i
d
de
n
layer
,
seco
nd
hi
d
d
en
l
a
y
er,
l
e
arn
i
n
g
ra
te
a
nd
mom
e
nt
um
a
re
s
electe
d
b
y
p
erfor
m
ing
se
ns
it
ivi
t
y
a
na
l
y
sis
w
h
e
r
e
the
selec
t
e
d
va
l
ues
of le
a
rni
n
g ra
te a
nd is re
f
e
rring t
o
t
he
m
in
i
m
u
m
R
M
S
E
v
a
l
u
e
of
ou
tpu
t
.
Tab
l
e
3. Resu
l
t
s
of
ANN c
ons
i
d
e
r
in
g al
l the
bes
t
par
am
eter
s
A
N
N
O
utput
f
or
K
=
2
4
T
r
a
i
ning
set
s
5785
T
e
sting
se
ts
720
N
u
m
b
e
r
o
f
ne
ur
on
in
1
st
h
i
d
d
e
n
l
a
ye
r
20
N
u
m
b
e
r
o
f
ne
ur
on
in
2
nd
h
idd
e
n
l
a
ye
r
10
N
u
m
b
e
r
o
f
output
1
Lea
r
n
i
ng
r
a
te
0
.
3
Mo
men
t
u
m
0
.3
Tra
i
ning
f
unc
tion
Le
v
e
nb
e
r
g-
M
a
r
qua
rdt
T
r
a
i
ning
RM
SE
0
.
5270
Te
sting
R
M
SE
0
.
5301
Re
gre
s
sion
0.
9934
2
M
i
ni
m
u
m
MA
PE
0
.
2832
Ma
xim
u
m
MAP
E
13.
237
7
Me
a
n
M
A
P
E
4.
4217
Evaluation Warning : The document was created with Spire.PDF for Python.
Int J
P
o
w
E
l
e
c
&
D
ri S
yst
IS
S
N
:
2088-
86
94
Perf
om
a
n
ce
co
m
par
is
on o
f
ar
tif
ic
ia
l i
n
te
l
l
i
g
e
n
c
e
te
ch
n
i
ques
in sh
ort te
rm
… (
M
uh
am
m
a
d Mur
t
a
d
ha O
t
hm
a
n
)
2
153
F
i
gure
s
6
(
a
)
a
n
d
6(
b)
r
epr
e
sen
t
t
he
r
e
s
u
lt
of
f
or
eca
ste
d
P
V
out
p
u
t
c
u
rrent
v
er
sus
a
c
t
u
al
t
arge
t
e
d
v
a
l
u
es,
and
the
regre
ssio
n
o
f
o
u
tp
ut
r
e
s
u
l
ts
o
b
t
a
i
ne
d
dur
in
g
t
h
e
te
st
i
n
g
proce
dure
of
A
N
N
,
re
spe
c
t
i
ve
l
y
.
I
n
F
igure
6(a)
,
the
ac
t
u
a
l
t
arge
te
d
o
u
t
p
u
t
i
s
in
b
lue
co
lo
ur
a
nd
t
h
e
for
eca
ste
d
P
V
out
pu
t
cu
rr
ent
is
i
n
r
e
d
colo
ur.
The
forecaste
d
pat
t
ern
o
f
h
o
u
r
l
y
P
V
out
pu
t
cur
r
ent
is
a
lmos
t
the
sa
me
w
it
h
the
p
a
t
t
e
rn
o
f
ac
tua
l
t
arge
te
d
va
lue
s
a
t
c
e
rtai
n
ho
urs.
H
ow
eve
r
,
t
h
e
r
e
is
i
nc
ons
i
s
te
nc
y
w
i
t
h
s
eve
r
al
l
ar
ge
e
rror
i
n
the
variat
io
n
be
t
w
e
e
n
t
h
e
for
eca
ste
d
a
n
d
targe
t
e
d
P
V
outp
u
t
c
urr
e
nt f
o
r
the
n
ex
t 2
4
h
o
u
r
s
.
(a)
(b)
F
i
gure
6.
S
TP
CF
for
the
n
e
x
t
24-h
ours
usin
g A
N
N for
the
(a) for
eca
st
e
d
PV out
p
u
t
curr
ent ve
rs
us
a
c
t
u
a
l
t
a
rg
et
ed
v
alu
e
s,
(
b
)
r
e
g
ressio
n
of
f
ore
caste
d
versus a
ct
ua
l P
V
out
pu
t c
u
rrent
3.2.
R
an
d
o
m for
e
st
(
R
F
)
The
m
u
l
tip
le
t
i
m
e
lags
o
f
h
o
u
r
,
tem
p
era
t
ure,
i
rr
adia
nc
e
an
d
cur
r
e
n
t
a
r
e
u
s
e
d
a
s
t
h
e
i
n
p
u
t
d
a
t
a
o
f
R
F
.
The
t
r
ain
i
ng,
t
est
i
ng
and
va
li
d
a
ti
n
g
p
roc
e
sses
of
R
F
a
r
e
conduc
te
d
usi
n
g
th
e
i
n
p
u
t
da
ta
w
ith
m
ulti
p
l
e
tim
e
lags
of
K
=
2
4
h
our
s.
T
he
R
F
is
c
ond
uc
t
at
t
hr
ee
di
ffere
n
t
c
a
ses
of
1
,
5
a
n
d
1
0
num
ber
of
t
re
es
(
Ntre
es
)
a
nd
e
v
er
y
tree
c
onsi
s
t
i
ng
of
5
l
ea
ves.
T
he
s
elec
t
i
o
n
f
or
t
h
e
num
b
e
r
of
t
r
ees
i
s
base
d
on
t
he
f
ac
t
t
h
a
t
s
in
gle
tre
e
i
s
suffi
c
ien
t
to
e
st
ima
t
e
an
y
fu
nct
i
on.
T
he
r
e
fore,
tw
o
tree
s
w
i
l
l
p
ro
v
i
de mor
e
prec
i
s
i
o
n
in determ
i
nin
g
the
mi
nimum
e
rror
.
The
mean
s
quare
error
(MSE)
is
obta
i
ne
d
fro
m
the
t
r
ai
nin
g
p
roc
e
dure.
H
ow
ever
,
i
n
t
he te
s
tin
g
proce
d
u
r
e,
t
he
M
S
E i
s
a
uto
m
at
i
call
y
c
o
m
p
a
red
i
t
s
ou
t
p
u
t
wi
t
h
th
e t
a
rg
et
ed
d
at
a
at
eac
h
lea
f
i
n e
v
ery
t
r
ee.
The
se c
ompa
rison
s
are
perfor
me
d
un
t
il
t
h
e
fi
ne
st
t
ree
s
e
xpa
nsio
n
is
a
chie
ve
d
g
i
v
i
n
g
t
h
e
minimum
ave
r
ag
e
of
R
MS
e
rror
f
o
r
the
fi
na
l
no
de
l
e
a
f
o
f
all
tree
s.
T
he
numbe
r
of
t
r
ees
c
hose
n
f
or
t
h
e
sens
i
tiv
it
y
a
n
a
l
y
s
i
s
t
o
de
t
e
rm
ine
th
e
be
st
pred
ic
ti
on o
f
PV
out
pu
t
curr
ent w
i
th m
in
i
m
um
R
M
S
E va
lue
s
RF
is
s
h
o
w
n
i
n
T
ab
l
e
4
.
Ta
b
l
e
4.
Resu
lts
o
f RF
c
ons
id
erin
g a
l
l
the
be
st pa
r
am
ete
r
s
R
F
Out
p
u
t
f
o
r
K
=
24
T
r
a
i
ning
se
ts
5785
Te
sti
n
g
se
ts
720
Vali
d
a
tion s
e
ts
720
T
r
a
i
ning
fun
c
tion
B
ootst
r
a
pping
Output
F
unc
ti
on
Re
gr
e
ssion
Nu
m
b
er
o
f
t
r
ees
1
5
1
0
Nu
m
b
e
r
o
f
l
eav
es
5
5
5
T
r
a
i
ning
R
M
S
E
0
.
8725
0
.
8753
0.
8756
Te
st
ing
RM
SE
0
.
0476
0
.
0089
0.
0078
Re
gr
e
ssion
0
.
9996
7
0.
9999
3
0.
9999
4
M
i
ni
m
u
m
MA
PE
0
.
0061
0
.
0016
0.
0111
Ma
xi
m
u
m
MA
PE
2
.
0735
0
.
6109
0.
7953
Mea
n
M
AP
E
0.
2731
0
.
2286
0.
2000
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN: 2088-8694
Int J
P
o
w
El
e
c
&
D
ri S
yst
,
V
ol.
10,
N
o.
3
, S
e
p
2
0
1
9
:
214
8
– 2
156
2
154
F
i
gures
7
(a
)
an
d
7(
b)
r
e
p
resent
t
he
c
ompa
rat
i
ve
r
esults
a
n
d
r
eg
re
ssi
o
n
of
f
o
r
ec
asted
P
V
o
u
t
p
u
t
c
urrent
versus
a
ctua
l
ta
rge
t
e
d
v
al
ues
obta
i
ned
from
the
testing
proce
d
u
r
e
o
f
A
N
N
,
respec
tive
l
y.
I
t
c
a
n
be
obse
r
ve
d
tha
t
the
fore
caste
d
ho
url
y
P
V
out
put
c
ur
ren
t
p
ro
vi
de
a
v
ery
si
milar
v
a
r
i
a
t
i
o
n
w
i
t
h
m
i
n
i
m
u
m
e
r
r
o
r
a
s
c
o
m
p
a
r
e
d
t
o
t
h
e
ac
tu
al
t
arg
e
t
e
d
v
a
lu
e
s
p
a
t
te
rn
.
(a)
(b)
F
i
gur
e 7.
S
TPCF
f
or
t
he
ne
x
t
24-h
ours us
in
g RF
f
or
t
he
(a
)
fore
ca
st
e
d
PV o
u
tp
ut
cu
rre
n
t
v
e
r
su
s ac
tu
al
targe
t
e
d
va
l
ues
,
(b)
re
g
ressio
n
of
f
ore
caste
d
versus a
ct
ua
l P
V
out
pu
t c
u
rrent
3.3.
MA
P
E com
p
aris
on
b
etwe
en
t
h
e
p
erform
a
n
c
e
of AN
N
and
R
F
Tabl
es
3
an
d
4
h
a
v
e
sho
wn
the
r
es
u
l
t
s
o
f ANN
an
d
RF, r
esp
e
ct
iv
e
l
y
.
The
outp
u
t
resu
lt
i
s
ob
ta
ine
d
b
y
con
s
i
d
eri
ng
to
m
ult
i
p
l
e
t
i
me
l
a
g
s
K
=
24
ho
ur
s
dur
in
g
t
e
st
i
n
g
p
roc
e
d
u
r
e
for
both
tec
h
n
i
que
s.
T
he
t
est
i
n
g
proce
dure
for
bo
t
h
t
e
c
h
n
i
que
s
w
ith
m
ul
t
i
p
l
e
time
lags
i
s
fur
t
he
r
inve
s
t
i
g
a
t
ed
by
com
p
ar
i
ng
the
MA
P
E
r
esul
t
s
o
f
P
V
ou
t
p
u
t
c
u
r
r
e
nt
.
By
r
eferr
i
n
g
to
t
h
e
MAP
E res
u
l
t
s
o
f ANN
an
d
R
F
i
n
Tabl
e
5
,
i
t
c
an
b
e ob
se
rv
ed
th
a
t
t
h
e
ANN
mo
d
e
l
pro
duce
s
a
h
ig
her
MA
P
E
v
alue
o
f
5.08
36
%
,
i
n
con
t
rast
w
it
h
t
h
e
M
A
P
E
o
f
0
.
057
9%
d
e
t
e
r
m
i
ned
by
t
he
R
F
in
day
t
w
o.
I
t
i
s
p
erspic
uo
us
i
n
Ta
b
l
e
5
tha
t
t
h
e
R
F
provide
s
m
o
st
acc
urate
pre
d
ic
ti
on
w
i
t
h
t
he
m
inimum
a
ver
a
ge
MA
P
E
r
esult
s
i
n
for
eca
stin
g
the
P
V
outpu
t
c
u
r
r
ent
a
s
c
o
m
pa
red
t
o
t
h
e
A
N
N
.
I
t
is
obv
i
o
u
s
t
ha
t
ba
g
g
i
n
g
tech
n
i
q
u
e
impr
ove
s
the
trai
n
i
ng
a
n
d
te
sti
n
g
pr
ocesses
o
f
R
F
in
obta
i
nin
g
t
he
b
es
t
re
sults
w
i
t
h
m
i
n
i
m
u
m
e
r
ror
in
f
ore
cast
i
n
g
.
Th
is
i
m
p
lies
tha
t
t
he
A
N
N
i
s
far
mor
e
c
omplica
t
e
d
t
ha
n
the
RF
i
n
term
s
of
i
nter
pret
in
g
an
d
un
dersta
n
d
i
n
g the
w
e
i
g
h
t
, e
asy to
o
ver-fit
th
e
m
odel a
nd
unpr
e
d
ic
te
d
in i
t
s
p
erform
ance
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Int J
P
o
w
E
l
e
c
&
D
ri S
yst
IS
S
N
:
2088-
86
94
Perf
om
a
n
ce
co
m
par
is
on o
f
ar
tif
ic
ia
l i
n
te
l
l
i
g
e
n
c
e
te
ch
n
i
ques
in sh
ort te
rm
… (
M
uh
am
m
a
d Mur
t
a
d
ha O
t
hm
a
n
)
2
155
T
a
b
l
e 5.
MA
P
E of
f
ore
cas
ted
PV
outpu
t
curre
nt for
t
he ne
x
t
24-
h
o
u
rs ob
t
a
i
ne
d
from
the
A
N
N
a
nd RF
Da
y
AN
N
R
F
K
=
24
MAP
E
(%)
M
A
P
E (%)
1
5.
9144
0
.
1184
2
5.
0836
0
.
0579
3
3.
0745
0
.
3008
4
2.
9441
0
.
1622
5
2.
4173
0
.
1184
6
6.
0031
0
.
1219
7
3.
3752
0
.
7556
8
4.
7268
0
.
1420
9
2.
5261
0
.
0864
10
6
.
9916
0
.
2823
11
8
.
9894
0
.
7953
12
2
.
2936
0
.
0190
13
6
.
2087
0
.
0574
14
1
.
7835
0
.
2749
15
4
.
5219
0
.
3175
16
1
.
0799
0
.
0118
17
6
.
7759
0
.
2154
18
1
3.
237
7
0.
1462
19
1
1.
492
9
0.
0111
20
0
.
3173
0
.
0243
21
0
.
9236
0
.
2045
22
0
.
2832
0
.
0320
23
3
.
3776
0
.
0203
24
4
.
4665
0
.
4610
25
4
.
4561
0
.
0198
26
5
.
9024
0
.
1180
27
5
.
0569
0
.
3915
28
3
.
0791
0
.
3008
29
2
.
9465
0
.
2508
30
2
.
4028
0
.
1835
Av
er
a
g
e
MAP
E
4.
4217
0
.
2000
4.
CONCL
U
S
ION
The
a
p
p
lica
t
io
n
of
a
rtific
ia
l
neur
a
l
n
etw
o
r
k
(
A
N
N
)
a
nd
ra
nd
om
f
ore
s
t
(RF
)
w
ith
w
ave
l
et
d
e
n
oisi
ng
and
mul
t
i
p
l
e
t
i
m
e
l
a
gs
K
=24
in
p
erform
i
n
g
short
te
rm
phot
o
v
o
lta
ic
c
u
r
r
e
n
t
f
o
r
ec
as
tin
g
(
STPC
F)
h
a
s
b
een
di
sc
usse
d
ela
b
or
ate
l
y
i
n
t
h
i
s
paper
.
T
he
r
esul
ts
s
h
o
w
n
p
ro
ved
t
ha
t
the
mode
ls
p
ro
po
sed
for
the
c
a
se
s
t
u
d
y
h
a
v
e
the be
nefi
t of p
rov
i
d
i
ng ac
cur
a
t
e
r
esult of STP
CF
. H
o
w
e
ver
,
t
he
RF
m
e
tho
d
sh
o
w
n
the
i
mpor
ta
nt of cho
o
s
i
ng
the
acc
urate
nu
m
b
er
o
f
tree
a
nd
le
af
t
o
be
u
se
d
a
s
i
t
w
i
l
l
a
ffe
ct
t
h
e
p
erf
o
rma
n
ce
o
f
R
F
.
Th
e
re
sul
t
s
hown
th
a
t
the
RF
m
etho
d
able
t
o
for
eca
st
t
he
P
V
out
put
c
urr
e
nt
f
or
t
he
n
e
xt
24
hou
r
s
a
nd
pro
v
i
d
e
m
o
r
e
a
c
c
ura
t
e
r
e
sults
of
S
TP
CF
w
it
h
m
i
nim
u
m
err
o
r c
o
mpa
r
ed
t
o A
N
N
.
ACKNOW
LEDG
E
MEN
T
Th
is
r
ese
a
rc
h
w
a
s
sup
porte
d
by
t
he
L
o
n
g
-Te
r
m
Re
sear
c
h
G
r
a
nt
(
LR
G
S
),
M
i
n
i
s
t
r
y
of
E
du
ca
ti
on
Ma
lays
ia
f
or
t
h
e
p
rogr
am
t
i
t
le
d
"D
e
c
a
r
bon
i
s
a
t
i
o
n
of
G
rid
w
i
th
a
n
Opt
i
ma
l
Con
t
ro
ller
an
d
Ener
g
y
M
a
n
ag
em
en
t
f
o
r
E
n
e
r
g
y
S
t
o
r
a
g
e
S
y
s
t
e
m
i
n
M
i
c
r
o
g
r
i
d
A
p
p
l
i
c
a
t
i
o
n
s
"
w
i
t
h
p
r
o
j
e
c
t
c
ode
6
0
0
-IRMI
/L
RG
S
5/
3
(0
01
/20
1
9
).
T
he
aut
h
ors
w
oul
d
a
l
so
l
i
k
e
t
o
ackn
o
w
l
e
d
ge
T
he
I
nst
itu
te
o
f
Rese
arc
h
Ma
nagem
e
n
t
&
I
nno
va
tio
n
(I
RMI)
,
Uni
v
ersi
ti
T
e
k
no
l
ogi
M
A
R
A
(
U
iTM
)
,
S
h
ah
A
l
a
m,
S
elang
o
r
,
Mala
ys
ia
f
or
t
he
f
ac
il
it
i
e
s
prov
i
d
ed
t
o
s
u
p
p
o
rt
o
n
th
i
s
r
esea
rch.
REFE
RENCES
[1]
M
.
H
osen
uzzam
an,
N
.
A
.
Rah
i
m
,
J
.
S
e
lv
araj,
M
.
H
asanu
zzam
an,
A
.
B.
M
.
A.
M
al
ek,
an
d
A.
N
ahar,
“
G
l
o
b
a
l
pro
s
p
ects
,
p
ro
gress
,
p
o
l
i
c
ies,
a
n
d
e
nv
ironm
ent
a
l
im
p
act
o
f
sol
a
r
ph
oto
volt
a
ic
p
o
w
er
g
enerati
o
n
,
”
Ren
e
w. Su
st
ain.
En
erg
y
Rev.
,
vol.
41,
p
p
.
2
84
–297,
2
0
15.
[2]
J.
F
.
M
.
P
ess
a
n
h
a
an
d
N.
L
eo
n
,
“
F
o
recas
ti
ng
lon
g
-term
elect
ricit
y
d
em
an
d
in
t
he
r
esi
d
en
tial
s
ecto
r
,”
Pro
cedi
a
Com
p
ut
. Sci
.
,
vol.
5
5
,
n
o.
I
t
q
m
,
p
p
.
5
2
9
–
5
3
8
,
2
015.
[
3
]
G
.
R
.
T
.
E
s
t
e
v
e
s
,
B
.
Q
.
B
a
s
t
o
s
,
F
.
L
.
C
y
r
i
n
o
,
R
.
F
.
C
a
l
i
l
i
,
a
n
d
R
.
C.
S
ou
za
,
“Lo
n
g
t
e
rm
e
lectricit
y
f
orecast
:
A
sy
st
e
m
ati
c
r
eview
,
”
Pro
c
e
d
ia Co
mp
ut.
Sc
i
.
,
v
o
l
.
5
5, n
o.
I
t
q
m,
p
p
.
5
49
–5
58
, 20
1
5
.
[4]
M
.
M
.
O
t
hm
an,
and
I.
M
usiri
n
,
"Op
t
im
a
l
s
izi
ng
an
d
op
erat
io
nal
s
trateg
y
of
h
y
b
ri
d
renewab
l
e
ener
g
y
s
ystem
u
s
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN: 2088-8694
Int J
P
o
w
El
e
c
&
D
ri S
yst
,
V
ol.
10,
N
o.
3
, S
e
p
2
0
1
9
:
214
8
– 2
156
2
156
hom
e
r," In
IEE
E
20
10
4
t
h Inter
n
a
tio
nal P
o
wer En
gi
neeri
n
g
an
d
O
p
timiz
a
t
i
on Co
n
f
er
ence
(
P
EOCO
)
, 2
01
0
.
[
5
]
A
.
Q
a
i
s
,
M
.
M
.
O
t
h
m
a
n
,
N
.
K
h
a
m
i
s
,
a
n
d
I
.
M
u
s
i
r
i
n
,
"
O
p
t
i
m
a
l
sizin
g
a
n
d
o
peration
al
s
trateg
y
of
P
V
an
d
m
i
cro-
hy
dro
,
"
In
2
013
IE
EE 7
t
h Internat
io
nal Po
wer Eng
i
n
eerin
g
an
d
Op
ti
miz
a
t
i
o
n
Co
n
f
eren
ce
(
P
EOCO
)
,
pp.
7
14
-717,
20
13
.
[6]
M.
G
hazvi
n
i
,
A
.
A
bbas
p
our-Teh
rani
-F
ard
,
M
.
F
o
tuh
i
-Fi
r
uzabad
,
and
M.
M
.
Othm
an,
"Op
t
i
m
izin
g
s
i
ze
and
op
erat
io
n
of
h
yb
rid
energy
s
yst
e
ms,
"
I
n
20
13
I
EEE
7th
In
te
rn
at
io
na
l
Po
we
r
En
gine
e
r
in
g
an
d
Op
t
i
m
i
za
tion
Con
f
e
r
e
n
c
e
(P
E
O
C
O
)
,
p
p
.
489
-494
,
2
01
3.
[
7
]
S
.
M
.
H
.
W
.
D
a
wi
,
M
.
M
.
Ot
hman,
I
.
M
u
s
i
r
i
n
,
A.
A
.
M.
K
am
ar
u
zaman
,
A
.
M
.
Arrif
f
in,
a
n
d
N
.
A
.
S
a
li
m,
"
Gam
m
a
S
t
i
r
li
ng
E
ngin
e
f
o
r
a
S
m
a
ll
D
e
s
ig
n
of
R
enew
abl
e
R
eso
u
rce
M
o
d
e
l,
"
In
do
nes
i
an
Jour
na
l
o
f
Electri
cal E
n
g
i
neer
ing
an
d Co
mp
u
t
er
Scien
c
e
,
vo
l.
8
,
no.
2
,
pp
.
35
0-3
5
9
,
2
0
1
7
.
[8]
F
.
Z
ahari
,
M
. M
.
O
t
h
m
a
n
,
I
.
M
u
s
i
rin,
A
.
A.
M
. K
am
aruzam
an,
N.
A
.
Sa
lim
,
a
n
d
B
.
N
.
S
h
e
i
kh
R
a
h
i
m
ul
la
h,
"
De
sign
of
a
s
m
a
l
l
r
enewab
le
r
eso
u
rce
m
o
d
e
l
bas
e
d
on
t
h
e
s
tirlin
g
engin
e
w
i
th
a
lp
ha
a
nd
b
e
ta
c
on
f
i
gu
ra
tion
s
,
"
Indo
nes
i
a
n
Jou
r
n
a
l
o
f
El
ectr
i
cal E
n
g
i
neer
ing a
n
d
Comp
u
t
er
S
c
ien
c
e
, v
ol. 8
,
n
o
. 2
, p
p.
36
0
-
3
6
7
, 20
1
7
.
[9]
F.
W
ang,
Z
.
Mi
,
S.
S
u,
a
nd
H
.
Z
h
ao,
“Short
-
term
s
olar
i
rrad
ian
ce
f
o
recasti
ng
m
o
d
el
b
ased
on
artifici
a
l
neu
r
al
n
et
wo
rk
us
in
g
statisti
cal featu
r
e param
e
t
e
rs
,”
E
n
er
gies
, vo
l
. 5,
n
o
.
5,
pp
.
13
5
5
–
1
3
7
0
,
2
01
2.
[10]
C
.
Vo
yant,
G.
N
o
t
t
on,
S
.
K
a
log
i
ro
u,
M
.
-
L.
N
iv
et,
C.
P
a
o
li,
F.
M
ott
e,
a
n
d
A
.
F
o
u
i
llo
y,
"
M
a
ch
in
e
l
e
arni
ng
m
e
th
ods
f
o
r sol
a
r rad
i
ati
o
n
f
o
recastin
g
:
A
review
,
"
Ren
e
wa
bl
e En
erg
y
,
vol.
105
,
p
p.
569
-582,
201
7.
[11
]
M
.
Din
g
,
L.
W
a
n
g,
a
nd
R
.
Bi
,
“A
n
ANN
-
bas
e
d
ap
pro
ach
f
o
r
f
o
recas
ti
ng
t
h
e
p
ower
o
u
t
pu
t
o
f
photo
v
o
l
t
a
ic
s
ystem,”
Pr
oced
ia
En
viron.
Sci.
,
v
o
l.
1
1
,
no.
P
ART
C,
p
p
. 1
30
8–1
31
5,
2011.
[1
2]
P
.
Man
d
al
,
S
.
T
.
S.
M
adh
i
ra,
A.
U
l
h
a
qu
e,
J
.
Men
g
,
a
nd
R
.
L.
P
i
n
ed
a
,
“
F
o
recast
i
n
g
power
o
u
t
p
u
t
of
s
o
l
ar
p
h
o
t
ovo
lt
aic
sy
st
e
m
u
s
i
n
g
w
av
e
l
et
t
ran
s
f
o
rm
a
nd
artif
i
c
i
a
l
i
n
t
ell
i
gen
ce
t
ech
ni
qu
es,
”
Pr
ocedi
a
Comp
ut
. S
c
i.
,
vol.
12
,
n
o
.
91
5,
p
p
.
33
2–
33
7,
2
01
2.
[
1
3
]
I
.
A.
I
b
r
a
h
i
m
,
T
.
K
h
a
t
i
b
,
A.
M
o
h
a
m
e
d
,
a
n
d
W
.
E
l
m
e
n
r
e
i
c
h
,
“
M
odelin
g
of
t
he outp
u
t current
o
f
a
pho
tov
o
ltai
c
g
ri
d-
con
n
ect
ed s
ys
tem
u
s
in
g
ran
d
o
m
f
orests techn
i
q
u
e,”
Ener
gy E
x
pl
or.
Exp
l
oit.
,
vo
l. 3
6, no
.
1
,
pp
. 1
32
–1
4
8
, 20
1
8
.
[14]
L
. Brei
m
an,
“Random
f
o
rests,”
M
a
ch.
L
e
ar
n.
,
v
o
l.
45
, n
o.
1
, p
p.
5
–
3
2
, 20
0
1
.
[15
]
M
.
K
a
yri,
I
.
Ka
y
r
i,
a
nd
M.
T
.
Genco
g
l
u
,
“T
he
p
erf
o
rm
ance
c
omparison
o
f
M
ultip
l
e
L
i
n
ear
R
e
g
ressi
on,
Rando
m
F
o
res
t
a
nd
Artificial
N
eu
ral
N
e
tw
ork
by
u
s
in
g
p
h
ot
ovo
lt
aic
and
atm
o
spheric
dat
a
,
”
20
17
1
4
th Int.
Conf. En
g. M
od.
El
ectr. Sy
st.
EMES 2017
, p
p
. 1–
4
, 2
01
7.
[1
6]
M
. H. H.
Harun
, M.
M
.
Ot
h
man
,
a
nd
I
.
M
u
siri
n,
“
Sh
ort
term
l
o
ad
forecast
i
n
g
(
ST
LF
)
usin
g
artifi
cial
n
eural
netwo
r
k
based mul
tiple
l
ags
o
f
t
ime serie
s
,”
L
ect. No
t
e
s
Comp
ut
. S
c
i. (includ
i
n
g
S
u
b
s
er
.
L
ect.
Notes
A
r
ti
f.
Int
e
ll
. L
ect. Not
e
s
Bi
o
i
n
f
orma
tics)
, v
ol
.
55
07
LN
C
S,
n
o
.
P
A
R
T
2
,
p
p
.
4
45
–4
46
,
20
09
.
[1
7]
M
. H.
H
.
Haru
n, M
.
M
.
O
t
h
m
a
n
, and
I
. M
u
s
i
r
i
n
,
"
S
h
ort
term
l
oad
forecast
i
ng
(
S
T
L
F
)
using
artifi
c
ia
l
neur
al
n
et
work
bas
e
d
m
u
lt
iple
l
ags
an
d
s
t
ationary
time
seri
es
,"
I
n
2
010
4
t
h
In
t
e
rn
ation
a
l P
o
wer
En
gin
eeri
n
g
and
Op
timiz
a
tion
Con
f
er
ence (
P
EOCO)
,
pp
.
36
3-3
7
0
, 20
1
0
.
[1
8]
M
.
M
.
O
th
ma
n,
M
.
H.
H
.
Ha
ru
n,
a
n
d
I.
M
us
irin
,
"
F
o
r
e
c
a
s
ting
s
hort
term
e
l
ectric
lo
ad
b
ased
o
n
s
t
a
t
ion
a
ry
output
o
f
a
r
tific
ia
l
ne
u
r
a
l
n
e
t
work
c
on
side
ring
s
e
q
ue
n
t
ia
l
p
r
oc
e
ss
o
f
f
e
a
t
u
re
e
xtract
ion
m
e
t
h
ods
,
"
In
2012
IE
EE
Interna
tional
Po
wer
E
n
g
i
n
e
e
r
in
g and
Opti
m
i
zatio
n
Co
n
f
eren
ce
M
e
l
a
ka,
M
a
l
aysia
,
p
p.
485
-48
9
,
20
12.
[1
9]
M
.
M
.
O
thman
,
M
.
H
.
H
.
Haru
n
, N.
A
.
S
al
im,
an
d
M
.
L
.
O
t
h
m
a
n, "Seq
u
en
ti
a
l
p
ro
c
e
ss
o
f
feat
ure
ex
tracti
o
n
m
e
th
od
s
f
o
r
artif
i
c
i
al
n
eural
n
e
two
r
k
i
n
s
h
o
rt
t
erm
lo
ad
f
ore
cas
tin
g,
"
ARPN
Jo
urn
a
l
o
f
E
n
g
i
neerin
g a
nd
A
p
p
l
ied Sci
e
nces,
v
o
l
.
10
,
n
o.
19,
p
p
.
883
0-88
38
,
2
015
.
[20]
M
.
M.
O
t
h
m
a
n,
M
.
H
.
H
.
Harun,
a
nd
I
.
Mus
i
rin,
"
Short
t
e
rm
l
o
ad
f
o
r
ecast
i
ng
using
artifi
c
i
al
n
eural
n
e
tw
ork
wit
h
f
eatu
r
e
ext
r
acti
o
n m
e
tho
d
and s
t
ati
onary
o
utp
u
t
,
"
In
20
12
IEE
E
In
ter
natio
nal P
o
wer
En
gi
neeri
ng a
nd
Op
timiz
a
tion
Con
f
er
ence M
e
laka
,
M
a
la
ys
i
a
,
p
p
.
4
8
0
-
4
84
,
20
1
2
.
[21
]
J
.
S
.
A
rm
stro
ng
, “
Illusio
n
s
in
R
eg
res
s
i
on
An
alysis
@
u
pen
n
.acad
em
i
a
.edu
,”
v
ol.
2
0
12
, n
o. 3
, p
p
.
9
6
1
–
9
6
7
, 2
012
.
[2
2]
P
.-H.
C
h
i
a
n
g,
S
.
P.
V
.
Chi
l
uv
ur
i
,
S
.
De
y
,
a
n
d
T
.
Q
.
Ngu
y
e
n
,
“
F
o
r
ecasti
ng
o
f
S
o
l
a
r
P
hotov
oltai
c
S
ystem
P
o
wer
Gen
e
rati
on
Using
W
a
vel
e
t
D
eco
m
p
o
s
i
t
i
o
n
and
Bi
as-Com
p
e
nsat
e
d
R
a
n
dom
Forest,”
20
17
Ninth
An
nu
.
IEEE
Gr
een
T
echn
o
l
.
Conf.
,
pp
.
2
60
–26
6,
2017
.
[23]
V
.
Lo
B
rano,
G.
C
iul
l
a,
a
nd
M
.
Di
F
a
l
co,
“A
rt
ifi
c
i
a
l
N
e
ura
l
Netwo
r
ks
t
o
P
r
ed
ict
th
e
P
o
w
e
r
O
u
tput
o
f
a
P
V
P
an
e
l
,”
Int
.
J.
Ph
oto
e
nergy
, vo
l
. 20
1
4
, p
.
12
,
2
0
1
4
.
[24]
R
.
P
e
rez,
S
.
K
i
valo
v,
J
.
S
c
hl
emmer,
K
.
H
e
m
k
er
J
.
D.
R
en
né,
an
d
T
.
E.
H
o
f
f
,
“Val
idat
ion
of
S
hort
a
nd
M
edium
Ter
m
Op
erati
onal S
o
lar Radi
ati
o
n
F
o
recast
s
in th
e
U.S.,” no
. 2
010, 2
00
9
.
[2
5]
P
. Bacher, H
.
M
a
d
s
en
, and
H.
A.
Ni
e
lsen
,
“On
lin
e sho
r
t
-
ter
m
s
o
la
r
po
wer f
o
recas
ting,
”
So
l.
Ener
gy
,
v
o
l
.
83
, n
o.
1
0,
pp
.
1
77
2–1
78
3,
2
0
09.
Evaluation Warning : The document was created with Spire.PDF for Python.