TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 8, August 201
4, pp. 5926 ~ 5931
DOI: 10.115
9
1
/telkomni
ka.
v
12i8.442
2
5926
Re
cei
v
ed Se
ptem
ber 30, 2013; Revi
se
d March 14, 2
014; Accepte
d
April 2, 201
4
Short-te
rm Power Prediction of the Photovoltaic
System Based on QPSO-SVM
Yang Lei
1
*, Shiping Zhou
2
, Yongjun Xia
1
, Xin
Shu
1
1
State Grid Hubei El
ectric Po
w
e
r Res
earch I
n
stitute, Xu
do
n
g
Roa
d
No.2
27
, W
uhan, Chin
a
2
State Grid Hubei El
ectric Po
w
e
r Comp
an
y,
Xu
do
ng R
oad
No.17
5
, W
uha
n, Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: 3042
31
569
@
qq.com
A
b
st
r
a
ct
Short-term po
w
e
r
predicti
on of
the
ph
otovo
l
taic
system is one of
the effe
ctive mea
n
s to
reduc
e
the adv
erse
effects of phot
ov
oltaic
pow
er o
n
the gr
id.
Si
n
c
e the effici
enc
y of the traditi
o
nal s
upp
ort ve
ctor
mac
h
i
ne (SV
M) pred
ictio
n
meth
od
is
lo
w
,
this pap
er
pro
poses
the
SVM bas
ed
on th
e p
a
ra
meter
opti
m
i
z
at
ion
method
of q
u
a
n
tum p
a
rticle sw
arm opti
m
i
z
a
t
i
o
n (QPSO), and
then
ap
ply
int
o
the
pow
er s
h
ort-
term pre
d
icti
on
of the photov
oltaic syste
m
. After co
mp
arin
g and a
naly
z
i
n
g the pred
ictio
n
results of SVM
base
d
on thr
e
e opti
m
i
z
at
io
n
meth
ods, w
e
find that
the
QPSO-SVM method
has
b
e
tter precis
ion
a
n
d
stability, w
h
ich
provi
des refere
nce to foreca
st
gener
atio
n po
w
e
r of the photovolta
ic system.
Ke
y
w
ords
: photovoltaic system
, power prediction, SVM,
QPSO
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Develo
pment
of scien
ce
and technolo
g
y
leads to
an eno
rmo
u
s amount of energy
con
s
um
ption,
the deman
d
for energy is incre
a
sin
g
rapidly. But the tradition
al
fossil en
ergy
s
o
ur
ce
s ar
e
no
t r
e
ne
wa
b
l
e
.
T
o
find
a n
e
w
s
u
s
t
a
i
na
ble
develo
p
ment
way of
the e
nergy so
urce
is
importa
nt in the future. Solar ene
rgy ca
n meet
huma
n
need
s, pho
tovoltaic power gen
eratio
n
is
one of the main uses
o
f
solar e
nergy. In rec
ent
years, ph
otovoltaic po
wer gen
eratio
n is
developin
g
v
e
ry ra
pidly, b
u
t photovoltai
c
po
we
r va
ri
es a
s
th
e we
ather
ch
ang
e
s
, it is u
n
cert
ain
and cycli
c
al,
large
-
scale p
hotovoltaic p
o
we
r will
ma
ke great imp
a
ct on the p
hotovoltaic g
r
i
d
-
con
n
e
c
ted system. An accurate p
r
e
d
i
c
tion of t
he photovoltai
c
can effe
ctively alleviate this
probl
em.
Ho
w to
accu
rate
ly predi
ct the
output of
pho
tovoltaic p
o
wer g
ene
ratio
n
syste
m
atta
ches
great im
port
ance to ma
stering the
ru
nning
ch
ara
c
teristics of
p
hotovoltaic p
o
we
r
ge
ne
rat
i
on
system, and
also
to wea
k
ening
the ne
gative
infl
ue
n
c
e
of ph
otovo
l
taic p
o
wer g
eneration
system
for power grid, it has becom
e a more an
d mo
re important
subj
ect of the research
on
photovoltai
c
power.
It’s
diffic
u
lt to predic
t the power of photo
v
o
lt
aic sy
st
em
be
ca
us
e
there are so
ma
ny
factors that
a
ffect the po
wer
of photovo
l
taic system. No
w
ma
ny
m
e
thod
s a
r
e
wi
dely used i
n
t
h
e
photovoltai
c
power p
r
edi
ction su
ch a
s
time serie
s
predi
ction, artificial ne
ural netwo
rk,
grey
forecast,
sup
port ve
ctor
m
a
chi
ne
(SVM) an
d
so o
n
[
1
-6]. SVM al
gorithm
re
pla
c
e
s
expe
rie
n
c
e
minimization
prin
ciple
of the tra
d
itional
machine
le
a
r
ning th
eo
ry by stru
ctural
risk minimi
za
tion
prin
ciple.
Co
mpare to
oth
e
r al
go
rithms, SVM algo
rithm ta
ke m
o
re adva
n
tage
s on th
e forecast
accuracy, but
the para
m
et
ers
of SVM model h
a
ve
a gre
a
t impa
ct on the foreca
st a
ccu
ra
cy.
Paramete
r op
timization be
come
s on
e of
the most important
conte
n
t in the rese
arch of SVM [7-
8].
Referen
c
e [9]
uses supp
ort
vecto
r
ma
chi
ne
al
gorith
m
i
n
to foreca
stin
g the
output
power
of photovoltai
c
po
wer g
e
n
e
ration
syste
m
, and put
forward a con
c
eptio
n of photovoltaic po
wer
predi
ction
system base
d
on
the forecasti
ng algo
rithm
of SVM.
Refere
nce [
10] intro
d
u
c
e
s
the
we
b
search
algo
rithm, gen
etic
algorith
m
an
d pa
rticle
swarm optim
ization algo
ri
thm.
After
compa
r
ing
the forec
a
s
t
res
u
lt
s
of the three differen
t
para
m
eter o
p
timization
m
e
thod
s, we
find the fo
re
cast results
of parti
cle
swa
r
m o
p
timizati
on
sup
port vecto
r
machine
(
PSO-SVM) i
s
si
gnifica
ntly better than the o
t
her two meth
ods.
Referen
c
e [
11] sele
ct
s simila
r days by the indi
cators of m
a
ximum tem
peratu
r
e,
minimum te
mperature, m
a
ximum humi
d
ity, mi
nimum humidity and so on. Th
en fore
ca
sts
the
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Short-te
rm
Powe
r Predi
cti
on of the Pho
t
ovoltai
c
Syst
em
Based on
QPSO-SVM (Yang L
e
i)
5927
output po
we
r
of photovoltai
c
po
we
r
ge
ne
ration system
by
least squ
a
re
s supp
ort vector
m
a
chi
ne
(LS-SVM
). Proved
the
L
S
-SVM can
do b
e
tter
at
fore
ca
st a
c
curacy
than
neu
ral
net
work
algorith
m
and
the normal S
V
M algorithm
.
This a
r
ticle
rese
arche
s
th
e paramete
r
optim
izatio
n
method of qu
antum pa
rticl
e
swarm
optimizatio
n(QPSO), then
introdu
ce
s
QPSO-SVM
to the sh
ort-t
e
rm p
o
wer p
r
edi
ction of t
he
photovoltai
c
system. The
method is val
i
dated by
a p
hotovoltaic
system data o
f
a photovoltaic
power statio
n
in
Wu
han.
T
he
re
sults sh
ow th
at
QPS
O
-SVM can
do
b
e
tter
i
n
speed, preci
s
i
o
n
and sta
b
ility, which provi
des referen
c
e to the s
hort-term po
we
r predi
ction of
the photovoltaic
sy
st
em.
2.
The Quan
tu
m Particle Sw
a
r
m Qptimi
zatio
n
Algori
t
hm
QPSO alg
o
rit
h
m is a n
e
w PSO algo
rit
h
m combin
es quantu
m
ph
ysics theo
ry
and the
traditional
P
S
O algo
rithm
,
whi
c
h i
s
b
a
se
d o
n
the
qua
ntum p
h
y
sics the
o
ry, and
re
ga
rd
s the
particl
es of P
S
O follow th
e motion of q
uantum phy
si
cs. So we ca
n describe th
e movement
o
f
particl
es by
quantum p
h
ysics. QPSO-SVM is a
n
improve
d
SVM algorit
hm whi
c
h take
s
advantag
e of QPSO param
eter optimi
z
at
ion [12-1
5
].
Whe
n
de
scri
bes the p
a
rti
c
le
s of PSO
by
quantu
m
physi
cs, we regar
d all th
e
particl
es
are movin
g
a
r
oun
d an attraction p
o
tenti
a
l cente
r
, we
reco
rd it as
q
i
=
(
q
i,1
, q
i,2
,
q
i,3
, …,
q
i,
m
).
i
i
s
the num
be
r o
f
the pa
rticle
s,
m
is the dim
ensi
on
of the
mathemat
i
c
al
pro
b
lem. T
h
e coordinate
of
the attraction
potential cent
er is:
q
i,j
=
i,j
.
p
i,j
+
[1–
i,j
]
.
p
g,j
, j =
0,1, …,m
(1)
i,j
is a
ran
d
o
m
num
be
r u
n
i
f
ormly di
strib
u
ted o
n
[0,1],
p
i,j
is the
opti
m
al lo
cation
particl
es
ever re
ached,
p
g,j
is the optimal locatio
n
the gr
o
up of p
a
rticle
s eve
r
reached.
In the qua
ntu
m
spa
c
e th
e
wave fun
c
tion
is use
d
to
descri
be the
state of pa
rticles. The
modulu
s
-squ
ared val
ue of
the wave fu
nction
rep
r
e
s
ents the
pro
b
ability densit
y of particle
s
to
appe
ar in any
place of the
spa
c
e. The fo
rmula i
s
as fo
llows:
|
|
2
dxdydz
=
Qdxdy
d
z
= 1
(2)
Q
re
presents the proba
bility of a
particle to appe
a
r
in point (
x,
y,
z
). Set
a one-
dimen
s
ion
a
l
probl
em fo
r
example, a
ssume a
si
ngle
parti
cle
at p
o
int
x
in
a on
e
-
d
i
me
ns
io
na
l
spa
c
e.
E
s
tabl
ish a one
-dim
ensi
onal
pote
n
tial
well in the attraction p
o
tential cente
r
q
. By solving
the Schrodin
ger Equ
a
tion
we can
get t
he probability
density funct
i
on
Q
, and th
en cal
c
ul
ate the
positio
n of the particl
e by Monte
Ca
rlo
stocha
stic si
mulation, the
basi
c
evolutio
n equatio
n of this
particl
e in QP
SO algorithm
is as follo
ws:
x
=
q
± (
L
.
In
(1
/u
))/2
(3)
L
is the
ch
a
r
acte
ri
stic le
ngth of
the
one-dime
nsio
nal pote
n
tial
well,
u
is
a r
a
nd
o
m
numbe
r u
n
ifo
r
mly dist
ribut
ed on
[0,1]. For a
m
-dim
en
sion
al spa
c
e,
we
can
a
s
su
me the attract
i
on
potential cent
er
i
s
q
i
=
(
q
i,1
, q
i,2
,
q
i,3
, …,
q
i,
m
), and e
s
tablish a
one
-dimen
sion
al
potential
well
for
every attra
c
ti
on p
o
tential
center i
n
e
a
ch
dimen
s
ion.
We can
define
a be
st ave
r
ag
e lo
cation
as
P
= (
P
1
, P
2
, P
3
,
…
,P
m
), for
n
p
a
rticle
s in a
m-dime
nsi
o
n
a
l spa
c
e, the
best average
locatio
n
is:
P
= [
P
1
(
t
),
…
,
P
m
(
t
)] =
[
p
i,1
(
t
),
…
,
p
i,
m
(
t
)]/
n
(4)
Then the
cha
r
acte
ri
stic len
g
th of the m-dimen
s
ion
a
l potential well can b
e
de
scri
bed a
s
:
L
i,j
=
2
|
P
j
–
x
i,j
|
(5)
is the co
ntraction
-
expa
n
s
ion
coeffici
e
n
t, in this pa
per
is lin
ea
rly decli
ning f
r
om 1 to
0.5. Combini
ng the (3
) an
d (5), we ca
n finally
get the evolution eq
uation of QP
SO algorithm:
x
i,j
(
t+1
)=
q
i,j
±
p
j
–
x
i,j
(
t
)
In
(1/
u
i,j
)
(6)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 592
6 –
5931
5928
3.
Short-term P
o
w
e
r Predic
tion of the Ph
otov
oltaic Sy
stem Base
d on the QP
SO-SVM
3.1. Data
Nor
m
alization
The origi
nal power
data
cannot be use
d
to pr
e
d
ict t
he o
u
tput p
o
w
er di
re
ctly becau
se
some
data is
error o
r
missi
ng. We shoul
d add the
mi
ssing d
a
ta and
corre
c
t the si
gnifica
nt errors
in the data by the adjacent data. Then n
o
rm
ali
z
e the
data by the equation a
s
fol
l
ows:
x
i
*
= (
x
i
–
x
min
)/(
x
mi
n
–
x
min
)
(
7
)
x
i
*
is the
normalize
d
data,
x
i
is th
e o
r
ig
inal data,
x
ma
x
is the maxi
mum data,
x
min
is
the
minimum dat
a.
3.2.
Input and O
u
tpu
t
Da
ta o
f
Prediction
The d
a
ta of
this
pap
er i
s
according
to
a ph
otovoltai
c
p
o
wer
stati
on in
Wuha
n
.
Before
forecastin
g we sh
ould
sele
ct simila
r d
a
ys for
photov
ol
taic po
we
r p
r
edictio
n. In this pa
per
we
u
s
e
grey relation
analysi
s
to
analyze the
climate
simil
a
rity of hi
storical d
a
ys
an
d fore
ca
st d
a
y in
terms of p
h
o
t
ovoltaic p
o
wer p
r
e
d
ictio
n
[16-1
9
], first
cal
c
ul
ating t
he
simila
r d
egre
e
of
ea
ch
climate in
dica
tors
between
the histo
r
i
c
al
days
a
nd fo
reca
st day, th
en have
a
weighted
sum.
At
last, sele
ct si
x most similar days and re
g
a
rd them a
s
the histo
r
ical days.
The input an
d
output data o
f
predictio
n is as follows:
Input data are the output
power data o
f
the historica
l
at each corresp
ondi
ng times an
d
its ne
arby times; th
e d
a
ta of
som
e
cli
m
ate in
di
cato
rs such a
s
te
mperature,
h
u
midity and
solar
radiatio
n. Output data are t
he output po
wer d
a
ta forec
a
s
t
ing by SVM of the forecas
t
day.
3.3. Parameter
Setting
The main poi
nt of SVM predictio
n is pa
rame
te
r opti
m
ization. Th
e
r
e are three i
m
porta
nt
para
m
eters
o
f
nonlin
ear su
pport ve
cto
r
r
egre
s
sion
ma
chin
e: pe
nalt
y
factor
C
,
n
o
n
-s
en
sit
i
v
e
lo
ss
coeffici
ent
and the
ke
rn
el width
coef
ficient of th
e
Gau
ssi
an ra
dial ba
sis
ke
rnel fun
c
tion
.
Since the
pre
d
iction i
s
a th
ree
-
dime
nsi
o
nal problem i
n
QPSO-SV
M
, we set the
three p
a
ra
m
e
ters
as
follows
:
C
= |
p
i,1
|,
C
> 0
(8)
= |
p
i,2
|,
> 0
(9)
= |
p
i,2
|, 0
<
< 1
(10)
3.4. Prediction
P
r
oces
s
The sp
ecifi
c
calcul
ation ste
p
s are as foll
ows:
(1) S
e
t the
si
ze of
parti
cle
s
n =
20
, the
dimen
s
ion
m =
3
, and
the l
o
catio
n
of the
numb
e
r
i
particl
e
in space
is
x
i
=
(
x
i,1
, x
i,2
,
x
i,3
, …
,x
i,
m
),
p
i
=
(
p
i,1
, p
i,
2
,
p
i,
3
,…,
p
i,
m
)is the
optimal lo
cati
o
n
particl
es ever rea
c
h
ed,
p
g
=
(
p
g,1
, p
g,2
,
p
g,3
, …,p
g,
m
) is the o
p
timal l
o
catio
n
the
g
r
oup
of p
a
rticles
ever re
ached,
t
is the number of cu
rrent iteration.
(2) Th
en initialize the lo
ca
tion of all particles
rand
om
ly, calculate the fitness val
ue and
the location
of each pa
rticle. In this paper, t
he fitness value is the RMSE of the predictio
n
results. Store
the fitne
s
s value
and th
e l
o
catio
n
of e
a
c
h
parti
cle i
n
function
pbe
st
, and the fitn
ess
value and the
location of th
e grou
p in fun
c
tion
gbe
st
.
(3)
Upd
a
te the locatio
n
of each parti
cle
by the functio
n
s a
s
follows:
q
i,j
(
t+1
)
=
i,j
(
t
)
.
p
i,j
(
t
)
+
[
1
–
i,
j
(
t
)]
.
p
g,
j
(
t
)
(11)
(
t+1
) =
max
–
t
.
(
max
–
mi
n
) /
max
(12)
x
i,j
(
t+1
) =
q
i,j
(
t
)
±
(
t
)
.
|
P
j
(
t
) –
x
i,j
(
t
) |
.
In
(1/
u
i,j
(
t
))
(13)
Cal
c
ulate the
fitness valu
e and the l
o
cation
of e
a
ch parti
cle aft
e
r up
dating,
and then
update the fu
nction
pbe
st
and
gbe
st
with the cu
rre
nt best lo
cation.
(4)
Ju
dge th
e re
sults, if the re
sult
s sa
tisf
y the term
ination
condit
i
on, stop
upd
ate and
output the bes
t
res
u
lt. Or return to s
t
ep 3.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Short-te
rm
Powe
r Predi
cti
on of the Pho
t
ovoltai
c
Syst
em
Based on
QPSO-SVM (Yang L
e
i)
5929
(5) After g
e
tting the
ap
propriate
pa
ra
meters
, input them in SV
M for the forec
a
s
t
ing
p
r
oc
es
s
.
The flow
cha
r
t is sho
w
ed in
Figure 1.
Figure 1. The
Predictio
n Proce
s
s of QPSO-SVM
4. Numerical
E
xample
In this pape
r the data is from a ph
oto
v
olta
ic po
wer station in Wuhan, the m
onitorin
g
interval is 10
min. We take
the data of six histor
ical days in whi
c
h the climate
co
ndition
s is most
simila
r to the
fore
cast
day
for hi
stori
c
al
data, to
fore
ca
st the outp
u
t power of t
he fore
ca
st d
a
y.
Code the GA
-SVM, SPSO-SVM and QPSO-SVM by MA
TLAB. Forecast the output power of the
photovoltai
c
system
by th
e thre
e pa
ra
meter
optimi
z
ation
metho
d
s a
s
above,
and th
en d
o
a
comp
ari
s
o
n
o
f
them. The result
s of pred
iction is
sho
w
ed in Table 1,
the predi
ctio
n step is 1
0
m
i
n.
Figure 2. The
Predictio
n Result
s of Thre
e Method
s
The
figu
re of the
predi
ction
re
sults shows that the
pre
d
ic
tion
re
sult
s of QPSO
-S
VM are
the mo
st simi
lar to the
a
c
tual po
we
r, a
nd the
GA-S
VM have the
least
simila
r
result. Analyzing
the results i
n
Table
1,
we
can find
that t
he n
u
mb
e
r
of the fo
re
ca
sting time
in
wh
ich th
e M
R
E i
s
0
10
20
30
40
50
60
70
0
1
2
3
4
5
6
7
8
9
7:
0
0am
-
1
8
:
00
p
m
(
T
h
e
I
n
t
e
r
v
al
o
f
P
r
e
d
i
c
t
i
on
i
s
10
m
i
n
)
O
u
tp
ut
P
o
w
e
r
(
M
W
)
A
c
tu
a
l
P
o
w
e
r
QP
S
O
-
S
V
M
SP
S
O
-
S
VM
GA
-
S
V
M
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 592
6 –
5931
5930
below 20% is 53 of the
QPSO-SVM
method,
and
the number
of SPSO-SVM is 49, 45
MRE
results of the
GA-SVM is u
nder 2
0
%
Cal
c
ulate the
averag
e rel
a
tive erro
r, me
an
ab
solute
error m
ean
square e
rro
r a
nd ro
ot
mean squa
re
error of ea
ch
method a
c
cording to the predictio
n re
sul
t
s.
Table 1. The
Predi
ction Errors
Prediction Methods
The Prediction E
rrors
MAE
(MW
)
MRE
(%)
MSE
(MW
)
2
RMSE
(MW
)
QPSO-SVM
1.5668
13.88
5.3045
2.3031
SPSO-SVM
1.7639
15.04
6.2389
2.4978
GA-SVM
2.3283
19.51
10.9117
3.3033
We
can
see
clea
rly fro
m
the results sho
w
ed
a
bove. Wh
en
we
have p
a
ram
e
ter
optimizatio
n
by the thre
e
ways, QPS
O
algo
rith
m ca
n find bette
r
para
m
eters
a
nd the p
r
e
d
ictio
n
results of QP
SO-SVM are
more
simila
r to the actual d
a
ta.
By analysis t
he fore
ca
stin
g re
sults
we
can find
th
at the rel
a
tive error of p
r
edi
cti
on in two
perio
ds is large. It’s b
e
ca
use
there
were
many
flu
c
tuation
s
of
the climate condition
s
in
the
forecast
day i
n
fact,
so th
e
output
powe
r
of the
ph
oto
v
oltaic p
o
we
r station
fluctu
ated in th
e two
perio
ds, but the fore
ca
stin
g curve
s
a
r
e
usu
a
lly smoot
h.
Overall, the
averag
e rel
a
tive erro
r of th
ree paramet
er
optimi
z
at
io
n method
s is below
20%, they a
ll have cert
ain ind
u
stri
al
referen
c
e v
a
lue
s
. But the QPSO
-SVM take
s m
o
re
advantag
es i
n
a
c
cura
cy.
And be
ca
use
the
state of
motion
of th
e pa
rticle
s i
n
QPSO i
s
o
n
ly
descri
bed
by
displ
a
cement
, the mod
e
l o
f
QPSO-SVM
is
simpl
e
r th
an the
othe
r t
w
o m
e
thod
s,
so
comp
utationa
l co
mplexity
and
com
putat
ional
sp
eed
o
f
QPSO-SVM
is
also b
e
tter than th
e oth
e
r
two method
s.
5. Conclu
sion
This
pap
er
u
s
e
s
QPSO
a
l
gorithm
for
para
m
eter o
p
timization
o
f
SVM, and
applie
s
QPSO-SVM i
n
to the
sh
ort
-
term
po
we
r
predi
ction
of
the ph
otovoltaic
syste
m
.
The te
st d
a
ta
is
from a photovoltaic power station in Wuhan, a
fter compari
ng with
GA-SVM and SPSO-SVM, we
find the
QPSO-SVM al
go
rithm can
do
b
e
tter at
ac
cu
racy a
nd
co
m
putational
sp
eed. T
he
re
sults
certify that
QPSO-SVM has feas
ibility and good perf
or
mance in the short-te
rm
power
predi
ction
of the photovoltaic sy
stem.
Referen
ces
[1]
Lu J, Qu HQ,
Liu C. T
he statistic of photovo
l
taic po
w
e
r pre
d
ictio
n
.
Nroth Chin
a Electric
Pow
e
r.
2010
;
38(4): 56
3-5
6
7
.
[2]
Lan
ge
M. Ana
l
ysis
of th
e
unc
ertaint
y
of p
o
w
e
r pr
edicti
o
n
in
dustr
y
.
Olde
nb
urg, Germa
n
y
:
Univers
i
t
y
o
f
Olden
burg. 2
0
03.
[3]
Joens
en AK,
Giebe
l G, La
ndb
erg L.
M
o
del
outp
u
t statistics app
lie
d
to w
i
nd pow
er pre
d
ictio
n
.
Procee
din
g
s
of Euro
pea
n W
i
nd E
nerg
y
Co
nferenc
e o
n
W
i
nd E
ner
g
y
for
the N
e
xt Mil
l
e
nni
um, Nic
e,
F
r
ance. 199
9: 117
7-11
80
.
[4] Niu
DX
,
Ca
o SH, Lu JC. Po
w
e
r loa
d
forecas
t
ing techn
o
lo
g
y
and its app
lica
t
ion. Beij
ing, C
h
in
a: Chin
a
electric p
o
w
e
r.
200
9
.
[5]
Pan DF
, Li
u H
,
Li YF
. A
w
i
n
d
spe
ed forec
a
st optimizati
o
n
model
bas
ed
on time ser
i
es
ana
l
y
sis a
n
d
Kalma
n
filter al
gorithm.
Power
System
Technology.
200
8; 32(7): 82-8
6
.
[6]
Vapn
ik V, Statistical le
arni
ng theor
y.
Ne
w
Y
o
rk, American. John W
i
l
e
y&So
ns. 1998.
[7]
Song Q, W
a
n
g
AM, Z
hang
YS.
T
he comb
inati
on pr
edicti
on of BT
P in sinteri
ng pr
oce
ss base
d
o
n
Ba
yesi
an fram
e
w
ork a
nd LS-
SVM.
T
E
LKOMNIKA Indone
sian Jo
urna
l
of Electrical En
gi
neer
ing
. 20
13
;
11(8): 46
16-
46
26.
[8]
Z
hang
XF
, Z
hao Y. Appl
icati
on of sup
port
vector
machi
n
e to relia
bi
lit
y
ana
l
y
sis of en
gin
e
s
y
stems
,
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2013; 1
1
(7): 3
552-
356
0.
[9]
Li R, Li GM. Photovolta
ic po
w
e
r pred
iction b
a
sed o
n
SVM.
Electric Power.
2008; 4
1
(2): 7
4
-78.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Short-te
rm
Powe
r Predi
cti
on of the Pho
t
ovoltai
c
Syst
em
Based on
QPSO-SVM (Yang L
e
i)
5931
[10]
Li J.
Rese
arch
on
p
a
rameter
optim
izatio
n
o
f
SVM. Wuha
n
,
Chi
na: C
entr
a
l
Chi
n
a
Norm
al
Univ
ersit
y
.
201
1
.
[11]
F
u
MP, Ma H
W
, Mao JR. S
hort-term ph
otovolta
ic p
o
w
e
r
pred
iction
bas
e
d
on
LS-SVM.
Power System
Protection a
nd
Contro
l.
201
2; 40(6): 65-
69.
[12]
Z
heng
K. R
eac
tive p
o
w
e
r
o
p
ti
mizatio
n
of
po
w
e
r s
y
stem
ba
sed
on
improv
ed QPSO a
l
g
o
r
i
thm. Sich
uan
,
Chin
a: South
w
est Jiaoton
g U
n
iversit
y
. 20
10
.
[13]
Sun J, F
eng B, Xu W
B
.
Particle sw
arm optimi
z
a
t
i
on
w
i
th particles
havi
ng qu
antu
m
be
hav
ior
.
Procee
din
g
of Con
g
ress on E
v
oluti
onar
y C
o
m
putatio
n, Portlan
d
, American
. 2004: 32
6-33
1
.
[14]
Sun J, F
a
n
g
W
,
Xu
XJ. An
op
timizatio
n
meth
od
of QPSO. B
e
iji
ng, C
h
i
na: T
s
ing
hua
Un
iver
sit
y
Pr
ess,
201
1
.
[15]
Yang
BJ, H
a
i
XY.
Rese
arch
on s
e
lecti
n
g
a
ddres
s
a
nd c
a
pacit
y
of DGs
i
n
d
i
stributi
o
n
n
e
t
w
o
r
k
bas
ed
on QPSO.
Shaanxi El
ectric Pow
e
r.
2010; 11
: 24-27.
[16]
Liu
YL, S
u
n
Y
C
, San
g
J
R
. R
e
searc
h
on t
h
e
fa
cts that
hav
e im
pact
on
th
e p
hotov
oltaic
po
w
e
r.
Wa
te
r
Resources and Power
. 2011; 29(1
2
): 200-
20
2.
[17]
W
R
, Z
hang B
M
, Sun HB. T
he res
earc
h
of
similar
da
ys
i
n
short-term p
o
w
e
r
pred
ictio
n
.
Journ
a
l
of
T
s
ingh
ua Un
iv
ersity
. 2004; 4
4
(1): 106-
10
9.
[18]
Yang Z
L
, T
i
an Y, Z
hang GT
.
Nonl
in
ear the
o
r
etical
fou
n
d
a
ti
on an
d impr
ov
ement of simil
a
r da
ys metho
d
for short-term load forec
a
stin
g.
Power System
Technology.
2006; 3
0
(6): 6
3
-66.
[19]
Li CB, Li
XH,
Z
hang R. Meth
od to sel
e
ct
si
milar d
a
y
s for short-term loa
d
forecasti
ng.
Autom
a
tion of
Electric Power System
.
20
08; 32(9): 69-
73.
Evaluation Warning : The document was created with Spire.PDF for Python.