TELKOM
NIKA
, Vol.11, No
.1, Janua
ry 2013, pp. 468
~47
5
ISSN: 2302-4
046
468
Re
cei
v
ed Se
ptem
ber 2, 2012; Re
vi
sed
De
cem
ber 5,
2012; Accept
ed De
cem
b
e
r
13, 2012
Cloud Particle Swarm Algorithm Improvement and
Application in Reactive Power Optimization
Hong
sheng Su
Dept. of Electrical En
gin
eeri
n
g, Lanz
h
ou Ji
a
o
tong U
n
iv
ersity, La
nzho
u, Ch
ina
88 W
e
st Annin
g
Roa
d
, Lanz
h
ou 73
00
70, Chi
n
a
e-mail: shs
e
n
@
16
3.com
A
b
st
r
a
ct
T
o
resolve th
e
probl
e
m
s that
cloud
particl
e
sw
arm opti
m
i
z
a
t
i
on (CPSO)
w
a
s easily trapp
ed i
n
local
mi
ni
mu
m
and p
o
ssesse
d
slow
converge
nce spe
ed a
n
d
early-maturi
ng
during d
i
stribu
tion grid re
activ
e
pow
er opti
m
i
z
ation, CPSO
a
l
gorit
hm
w
a
s i
m
pr
ove
d
bas
e
d
on c
l
o
ud
di
gital fe
atures
i
n
this p
a
p
e
r. T
he
meth
od co
mb
i
ned Loc
al sear
ch w
i
th global search
togeth
e
r
based on sol
u
tion spac
e transform, w
here the
crossover an
d mutati
on o
pera
t
ion of the parti
cles
w
e
re imp
l
emente
d
base
d
on nor
ma
l cloud o
perator.
An
d
thus, the defects of CPSO algori
thm
were better tackled. Finally,
applied in bus IEEE30 system
, the
simulati
on res
u
lts show
that the better gl
oba
l soluti
on
is attai
ned us
in
g the i
m
pr
ove
d
CPS
O
algorith
m
, a
n
d
its converg
enc
e spee
d an
d a
ccuracy poss
e
sses a dra
m
ati
c
impr
ove
m
ent
.
Key
w
ords
:
cl
oud pa
rticle
swarm
,
Im
provem
ent, distrib
u
tion grid, rea
c
tive p
o
wer o
p
tim
i
zation
Copyrig
h
t
©
2013
Univer
sitas Ahmad
Dahlan. All rights res
e
rv
ed.
1. Introduction
Rea
c
tive po
wer optimi
z
ation is
a
mult
i-con
s
traint, large-scale a
nd
nonlin
ear
combi
nato
r
ial
optimization
problem in
power sy
st
e
m
s. It is a r
eactive re
gul
atory method
to
acq
u
ire on
e or more syst
em optimizati
on aims
thro
ugh the cont
rol of some variabl
es un
de
r all
the con
s
traint
s when the
system pa
ram
e
ters
and th
e
loads
are
gi
ven beforeha
nd [1]-[3]. The
mean
s to resolve the
probl
em
s in
clud
e co
nve
n
tionally cla
ssi
cal al
go
rithm and
arti
ficial
intelligen
ce
method
s. Th
e cla
s
sical al
gorithm
s in
cl
ude lin
ear
re
gulation m
e
th
od, and
non l
i
near
regul
ation, an
d as well as
mixed intege
r prog
rammi
n
g
and so on, who
s
e mai
n
probl
em
s are
the
disp
osal of discrete varia
b
les an
d multi-extrem
um
sea
r
ching. In process of optimizatio
n the
gene
ral pu
zzl
e
s such as di
mensi
onality curse a
nd la
rger calculatio
n error exist
s
, as a re
sult, the
ideal optimi
z
ation aims a
r
e hard to be
acq
u
ire
d
[4]-[6]. And artificial intelligen
ce method
s such
as si
mulated
anneali
ng, geneti
c
algo
rithm, immun
e
algorith
m
, ant colo
ny algorithm, pa
rticle
swarm
optimi
z
ation, a
nd
n
eural
net
work have
be
en
widely
appli
ed to resolve
power
syste
m
s
rea
c
tive po
wer optimi
z
atio
n [7]-[9]. The
s
e al
gorith
m
s with swa
r
m i
n
telligen
ce
b
a
se
d have
be
tter
global
sea
r
ch
ing ability and
process the
discrete
m
u
lti-obje
c
tive opt
imization p
r
o
b
lems. Howe
ver,
singl
e algorit
hm gene
rate
s many defe
c
ts like lo
ca
l
extremum an
d slow conve
r
gen
ce
spee
d so
that the desired re
sults i
s
difficult to be achiev
ed. In
rese
nt years, some ne
w rea
c
tive power
optimizatio
n
method
s, such as dynami
c
adaptive
diff
erential
evolu
t
ion algo
rithm
and im
prove
d
particl
e
swa
r
m optimizatio
n algo
rithm, a
r
e p
r
op
os
ed i
n
[10-11], re
spectively, wh
ere th
e sele
ction
of weight and convergence prec
i
s
ion are still not too accurate.
Based on it,
according to
the
descri
p
tion
s on clo
ud mod
e
l that it embodie
s
the ba
sic prin
cipl
es o
f
speci
e
s evol
ution in nature,
i.e., certainty
is contain
e
d
in un
cert
ain
t
y, and st
abil
i
ty is attache
d
to variatio
n in kno
w
led
ge
rep
r
e
s
entatio
n, cloud pa
rticle swa
r
m o
p
timizati
on
(CPSO) based
on clou
d digi
tal features (Ex,
En, He) i
s
i
n
vestigate
d
and a
pplie
d, the re
su
lt
s
indicate CP
SO ha
s bett
e
r
stability a
n
d
rand
omn
e
ss [12-1
3
]. To aim at the flaws of conv
e
n
tional PSO tha
t
the inertia weight gen
erati
n
g
mech
ani
sm can not refle
c
t the practi
cal
see
k
i
ng p
r
o
c
e
ss, CPSO
divides the whole pop
ulati
on
into three su
bpop
ulation
s
,
and diverse
generating st
rategi
es of
inertia weig
hts are ap
pli
ed.
CPSO wh
ose
the modulati
on strategie
s
of inerti
a wei
ghts ad
opt the X condition
cloud g
ene
ra
tor
is defined as self-ada
ptive CPSO. But
self-a
dapt
ive CPSO also e
x
pose so
me sho
r
tco
m
ing
s
in
power
syste
m
s rea
c
tive p
o
we
r optimi
z
ation such as
easy tra
ppin
g
in a lo
cal m
i
nimum, an
d early
mature an
d poor conve
r
g
ence pre
c
isi
o
n, and etc,
it is expected to impr
oved, further [14-1
5
].
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Clou
d Particl
e
Swarm
Algorithm
Im
provem
ent and Application … (Hon
gsheng S
u
)
469
Hen
c
e, in this pape
r CP
SO algorithm
is improved
from the two facets an
d
applied in b
u
s
IEEE30 syste
m
, which i
s
d
e
fined a
s
ICPSO algor
ith
m
, the excellent results a
r
e achi
eved a
n
d
indicate that the conve
r
ge
n
c
e sp
eed an
d
accu
ra
cy
of reactive po
we
r optimizatio
n
for distribution
grid are well solved.
2.
The Propos
e
d
Algorithm
2.1. Basic CPSO Algori
t
hm
To aim at
the defect of basic PSO algorithm
that the strategy
to
decre
ase prog
ressively
of inertia
wei
ght ca
n not re
flect the pract
i
cal
s
e
ar
ch
in
g p
r
oc
es
s
,
C
PSO
ma
ke
s
the
imp
r
o
v
emen
ts
on it, i.e., it
will divide the particle
s
into
the
three su
bpop
ulation
s
durin
g evolution, and diverse
updatin
g strat
egie
s
of inerti
a weig
hts ar
e
applied, the
model is d
e
scribed b
e
lo
w.
11
2
2
(1
)
(
)
(
)
(
)
()
(
)
tt
t
t
id
id
id
id
gd
id
vv
c
r
P
x
c
r
P
x
(1
)
)
1
(
)
(
)
1
(
t
id
t
id
t
id
v
x
x
(2)
2
'
2
)
(
2
)
(
*
5
.
0
~
9
.
0
En
Ex
fi
e
(3)
whe
r
e Eq.1 i
s
the speed
updatin
g formula, Eq.2 is displa
cem
e
n
t
updating formula, and Eq
.3 is
updatin
g formula of inerti
a weight.
V
id
is the flying speed of th
e
i
th partic
l
e in
D
-dime
n
sional
spa
c
e, an
d
x
id
is the displ
a
cem
ent of the
i
th particl
e in
D
-dim
en
sion
al spa
c
e,
and
is
the
updatin
g formula of in
erti
a wei
gh. In
(1
) to (3), p
a
ra
meter
t
exp
r
e
s
ses th
e itera
t
ion times, a
n
d
c
1
and
c
2
a
r
e n
on-n
egative l
earni
ng facto
r
s in rang
e of 1 to 2, and
r
1
and
r
2
are random va
ria
b
les
with a sco
pe
of zero to one
,
P
id
is the optimal po
sition
sou
ght by
the
ith particle
so far today, a
n
d
Pgd is the optimal position
sought by th
e whole parti
cle until now, fi is the
fitness value of the
ith
particl
e,
E
x
is the desi
r
ed v
a
lue an
d
E
n
is the entropy of cloud d
r
op
s.
In CPSO alg
o
rithm, the p
a
rticle
s a
r
e
di
vided into the
three
sub
p
o
pulation
s
a
ccordin
g to
their fitness
values in ea
ch it
eration p
r
ocess, and they evolve
a
c
cordi
ng to diverse evoluti
on
strategi
es. L
e
t
the colo
ny
scale of the
particl
e be
N
,
and the
n
the
averag
e fitness value
of the
particl
es can be
expre
s
sed
by
av
g
1
1
N
i
i
f
f
N
(4)
Let
'
av
g
f
be the averag
e fitness
value of the p
a
rticle
s wh
ose fitness valu
es are large
r
than
f
av
g
, and
''
av
g
f
be the averag
e fitness value of the
particle
s
who
s
e
fitne
s
s
value
s
a
r
e
lowe
r
tha
n
f
avg
, and the
fitness value
of
the most
optimal particle is expre
s
sed usi
ng
f
mi
n
. In
c
onc
rete
iteration p
r
o
c
ess, the inerti
a weig
ht upd
ates a
c
co
rdin
g to (3) b
u
t di
verse
evoluti
on strategie
s
are
applie
d. If
f
i
is
better than
'
av
g
f
, then the
evolution
strate
gy is
a
dopte
d
acco
rdin
g to
(1
), but the
se
con
d
item in the right si
d
e
in (1), i.e., co
g
n
itive model, is aba
nd
oned. Thi
s
m
ean
s the part
i
cle
evolves acco
rding to so
ci
al model alo
ne so as
to
accel
e
rate t
he global co
nverge
nce. An
d
conversely, if
f
i
is worse t
han
''
av
g
f
, then th
e evolution st
rategy is
ado
pted acco
rdin
g to (1), but
the third ite
m
in the rig
h
t
side in
(1),
i.e., so
cial
model, is i
g
n
o
red. T
h
is m
ean
s the pa
rticle
evolves a
c
co
rding
to cog
n
itive model
alone t
o
a
c
celerate the
conve
r
g
e
n
c
e sp
eed
of
the
particl
es that they possess poor
pe
rform
ance. Finally, if
f
’
avg
<
f
i
<
f
’’
avg
, then the evolution strate
gy
of the particle
s
ado
pts the full model like (1).
2.2. Impro
ved CPSO Al
gorithm
As previously
mentione
d, CPSO still ex
poses
som
e
deficie
nci
e
s d
u
ring
appli
c
at
ion, and
expect
s
to be improved. Hen
c
e, we m
a
ke t
he follo
wing imp
r
ove
m
ents for CP
SO using cl
o
u
d
digital feature
s
(Ex, En, He) to encode.
1) With the ai
m of populati
on altern
ative and
sol
u
tion
spa
c
e tra
n
sfo
r
m, the global
search
and lo
cal sea
r
ch a
r
e
combi
ned.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 468 – 4
7
5
470
The majo
rity of the operating time is
co
ns
um
ed in p
opulatio
n upd
ating in ba
sic CPSO
algorith
m
, and more
over, in the evolution later
pe
ri
od, the conv
erge
nce sp
e
ed is often more
slo
w
, and so popul
ation alt
e
rnative an
d solutio
n
sp
ace cha
nge a
r
e
introdu
ce
d.
The main thi
n
kin
g
of pop
ulation altern
ative
is that the wh
ole pa
rticle swarm is divided
into several sub
pop
ulatio
ns, one of which i
s
defin
ed as mai
n
popul
ation, and other o
n
e
s
is
defined a
s
a
u
xiliary popu
lations, they
see
k
the o
p
timizing ai
m
in solution
spa
c
e a
pplyi
ng
different see
k
ing mea
n
s. Duri
ng se
eki
ng pro
c
e
ss, the parts of main popul
ation and auxiliary
popul
ations a
r
e excha
nge
d to ensure t
he diversit
y of the particles in main populatio
n un
der
some
co
nditi
ons. In this
way, the early mature
can
b
e
avoided fo
r main colony
to guarantee
the
global extre
m
um to be foun
d.
In CPSO, the travers
a
l
s
pac
e is
[-1,
1] o
f
every dime
n
s
ion
of
t
he p
a
r
t
i
cle
s
.
Fo
r
co
rre
ct
ly
evaluating th
e sup
e
ri
ority-i
n
ferio
r
ity of the clo
ud
p
a
rti
c
le
s in curren
t position, the
solution
sp
a
c
e
transfo
rm i
s
required, that is, from unit spa
c
e
I
=[-1, 1]
n
to optimiza
t
ion solutio
n
spa
c
e ma
ppi
ng.
Let the ith cloud ope
rato
r of the particle
P
j
be [
i
j
,
i
j
]
n
, and so corre
s
p
ondin
g
solution sp
ace
variable
s
a
r
e
descri
bed by
1
[(
1
)
(
1
)
]
2
1
[(
1
)
(
1
)
]
2
jj
j
ic
i
i
i
i
j
jj
ii
i
i
i
Xb
a
Xb
a
(5)
And then th
e
optimizatio
n
is ma
de in th
e sol
u
tion
sp
ace. If the o
p
t
imal value is better tha
n
the
curre
n
t best solution, and repla
c
e it usin
g the optimal value.
2) To reali
z
e the cro
s
sover and mutation operatio
n of the
particle
s
acco
rdin
g to normal
clou
ds o
perator so a
s
to im
prove the see
k
ing fa
shio
n of the algorith
m
.
Cro
s
sove
r an
d mutation probabilitie
s are descri
bed a
s
follows.
'2
'2
fm
m
fm
fm
ma
x
1
2
'
()
'
2(
)
1
'
3
()
/
/
(,
)
f
fE
x
En
c
FF
Ex
f
x
x
FF
F
F
En
f
f
c
He
E
n
c
En
RA
N
D
N
E
n
H
e
ke
f
f
p
kf
f
(6)
whe
r
e
x
f
an
d
x
m
respe
c
tively express father individu
al
, mother individual,
F
f
and
F
m
resp
ectiv
e
ly
expre
ss fath
e
r
individual fitness
an
d mot
her individ
ual
fitness,
c
1
an
d
c
2
are
cont
rol variable
s
,
f
is avera
ge fitness.
Defini
tion 1(
mutation
):
T
o
give out the thre
shol
d
N
and
K
b
e
foreh
and, when the glo
b
a
l
extremum do
es not chan
g
e
, or who
s
e
chang
e ran
ge l
e
ss than
K
in contin
uou
s
N
times
iteratio
ns
durin
g evolution pro
c
e
ss, at the moment, the par
ticles are con
s
i
dere
d
to get into the local
extremum, a
c
cording
to gl
obal extremu
m
, all t
he p
a
rticles
are imp
l
emented
mut
a
tion op
eration
throug
h the n
o
rmal
clou
ds
gene
rato
r [12
]
.
Defini
tion 2
:
1-dim
e
n
s
ion
a
l normal clo
uds ope
rato
r, defined as A
r
Forward
C(Ex,
En,
He), is
a
mappin
g
of t
he whole
cha
r
acte
ri
stic fro
m
qualit
ative expre
ssi
on
to
quantit
ative expre
ssi
on,
i.
e.,
π
: C
→Π
, and
the following
con
d
ition
s
re
quire to be m
e
t.
22
{(
,
)
1
}
;
{(
)
,
1
}
;
{(
,
)
,
,
e
x
p
(
-
(
-
)
/
(
2
)
)
}
.
i
ii
i
ii
i
i
i
i
i
tN
o
r
m
E
n
H
e
i
N
Xx
N
o
r
m
E
x
t
t
i
N
xy
x
Χ
ty
x
E
x
t
,
,,
(7)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Clou
d Particl
e
Swarm
Algorithm
Im
provem
ent and Application … (Hon
gsheng S
u
)
471
whe
r
e Norm
is the normal
rando
m vari
able fun
c
tion,
and
t
i
is clo
ud dro
p
, N is the numbe
r o
f
clou
d dro
p
. Accordi
ng to (7
), the qualitati
v
e con
c
ept C(
Ex
,
En
,
He
) is co
nverse
d as a
cloud d
r
op
set expresse
d by nume
r
ical values
, a
n
d
that reali
z
e
s
the tra
n
sfo
r
mation from
con
c
e
p
t sp
ace to
nume
r
ical space. Obvio
u
sly, 1-dim
e
nsio
nal no
rmal clo
ud o
perato
r
may
expand int
o
n
-
dimen
s
ion.
Duri
ng CPS
O
evolution, some
ca
se
s oft
en appea
r such a
s
no
contem
po
rary
optimal
solutio
n
, and
that the more evolution
a
r
y devia
tes from the opti
m
al solutio
n
. To chan
ge
the
status, the fol
l
owin
g improv
ements a
r
e required.
As the col
o
n
y
initializatio
n, the initial
value is reco
rded
on the
curre
n
t positi
on and
velocity of ea
ch p
a
rticl
e
, a
nd then, the fi
tness
of ea
ch
particl
e is
cal
c
ulate
d
, and j
udge
wheth
e
r it
rea
c
he
s mut
a
tion thre
sh
ol
ds. If the con
d
itions a
r
e m
e
t, accordi
n
g
definition 1
each pa
rticle
is
impleme
n
ted mutation ope
ration, and otherwise acco
rd
ing to (1) a
nd (2) ea
ch p
a
rticle is carri
e
d
out updating
operation. Afte
r the end of
each ge
nera
t
ion, the opt
imal solutio
n
is sele
cted fro
m
the three su
bpop
ulation
s
,
and defined
as the global
optimal so
lution. If the
optimal soluti
o
n
meets th
e fitness
requi
re
ments, o
r
iteration times a
rrive at the
set times, the
evolution p
r
o
c
e
s
s
terminate
s
.
For the impro
v
ed algorithm
, the paramet
ers
Ex
,
En
,
He
,
K
, and
N
, and as well
as the
inertia wei
ght
and accel
e
rating facto
r
s
C
1
and
C
2
, have very important influen
ce
s on the
algorith
m
pro
pertie
s
. Obvi
ously, these improvem
ent
measures n
o
t only enhan
ce the diversity of
the colony, but also improve the
ability
to search opt
imization,
and reflect the
ability of normal
clou
d ope
rato
r to operate the parti
cle
s
.
3. Res
earc
h
Method
3.1. Reactiv
e Po
w
e
r Op
timization Mo
del Desc
ripti
on
Und
e
r th
e co
ndition of
po
wer sy
stems
rea
c
tive po
wer b
a
lan
c
e, p
o
we
r
system
s rea
c
tive
power
optimi
z
ation ta
ke
s t
he ge
ner
ator
bus volta
g
e
s
, and the
tran
sform
a
tion
rat
i
o of tran
sfo
r
mer
on-lo
ad volta
ge reg
u
lating
, and compe
n
satio
n
cap
a
c
itor a
s
cont
rol variable
s
, and re
du
ce
s the
grid l
o
ss a
nd
improve
s
the
quality of po
wer voltage
a
s
the
aim. In t
h
is p
ape
r, fro
m
co
nsi
deration
of econ
omi
c
asp
e
ct, activ
e
grid lo
ss m
i
nimum
is
sel
e
cted a
s
the
optimizin
g ai
m, and then,
the
aim function i
s
de
scribe
d a
s
follows[16]
G
m
a
x
m
in
m
a
x
m
in
2
2
v
lo
s
s
11
1
mi
n
iG
i
i
i
Gi
Gi
LM
N
ii
i
UQ
FP
UU
Q
Q
(8)
whe
r
e
U
i
and
Q
Gi
are ex
pre
s
sed by
mi
n
mi
n
mi
n
m
a
x
ma
x
ma
x
mi
n
m
i
n
mi
n
m
a
x
ma
x
max
,
0
,
,
0
ii
i
i
ii
i
i
ii
i
i
Gi
Gi
Gi
Gi
G
i
Gi
Gi
Gi
Gi
Gi
G
i
G
i
UU
U
U
UU
U
U
UU
U
U
QQ
Q
Q
Q
Q
Q
Q
QQ
Q
Q
And the con
s
t
r
aints of the
v
a
riabl
es a
r
e e
x
presse
d by
min
m
a
x
G
min
m
a
x
mi
n
m
a
x
T
min
m
a
x
G
mi
n
m
a
x
1
,
2,
,
1
,
2,
,
1
,
2
,
,
1,2
,
,
1
,
2,
,
Gi
Gi
Gi
ii
i
M
ii
i
Gi
Gi
Gi
c
i
ci
ci
c
QQ
Q
i
n
UU
U
i
n
QQ
Q
i
n
TT
T
i
n
UU
U
i
n
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 468 – 4
7
5
472
And the powe
r
con
s
trai
nts
are represent
ed by
1
PQ
1
(
c
os
sin
)
0
,
1,2,
,
,
.
(s
i
n
c
o
s
)
0
,
1
,
2
,
,
.
n
G
i
Li
i
j
ij
ij
ij
ij
j
n
G
i
Ci
L
i
i
j
ij
ij
ij
ij
j
PP
U
U
G
B
i
n
i
j
QQ
Q
U
U
G
B
i
n
whe
r
e the im
plicatio
n of each vari
able
are p
r
e
s
ente
d
belo
w
.
N
: Gene
rato
r node
s total n
u
mbe
r
.
M
: Load nod
e
s
total numbe
r.
L
: Grid circui
t number.
P
loss
:
Power
sys
tems
grid loss
.
U
i
,
U
i
max
,
U
i
min
: Node voltag
e, voltage limitation.
Q
Gi
,
Q
Gi
ma
x
,
Q
G
imin
: Generator re
active po
wer, rea
c
tive power limit.
v
,
G
: Cro
s
s-bord
e
r p
enalt
y
coefficient
s.
Q
Ci
:
Compe
n
s
at
ion
cap
a
cit
o
r ca
pa
cit
y
.
U
Gi
: Gene
rato
r terminal volt
age.
T
i
: Adjustable transf
o
rme
r
.
P
Gi
: Generato
r
active po
we
r.
G
ij
: Mutual condu
ctan
ce b
e
twee
n nod
e
i
and nod
e
j.
B
ij
: Mutual susceptan
ce b
e
t
ween b
u
s
i
a
nd
j.
δ
ij
: Voltage
pha
se differe
nce b
e
twe
en
node
i
an
d
j.
n
PQ
: PQ node numbe
r.
P
Li
,
Q
Li
: Load node
s a
c
tive power an
d re
active po
wer.
3.2. Distribution Grid Reactiv
e Po
w
e
r
Optimiza
ti
on Metho
d
Using Impro
v
ed CPSO
In CPSO alg
o
rithm, the co
uld dro
p
s a
r
e
generated u
s
ing X co
nditi
ons
clou
d ge
nerato
r
s,
that is cloud drop exp
r
e
s
sed by drop(x
0
,
μ
i) as the
particl
es, the displ
a
cement
s of the particle
s
in solution sp
ace corre
s
po
nds to co
ntrol
variables
in
power sy
ste
m
reactive po
wer o
p
timizat
i
on,
su
ch a
s
termi
nal voltage
U
G
of the generator, pa
rallel
capa
citor
ca
pability
Q
C
, and the tran
sf
orm
ratio
T
K
of transformer o
n
load voltage regulatin
g, t
he numbe
r of dimensi
o
n
s
of each parti
cle
is
equal to the n
u
mbe
r
of the control varia
b
l
es, that is,
T
KN
K
CN
C
GN
G
i
K
C
G
T
T
Q
Q
U
U
x
,
,
,
,
,
,
,
1
,
1
1
(9)
whe
r
e
N
G
,
N
C
, and
N
K
re
spe
c
tively expre
ss the n
u
m
ber of the
gene
rato
rs, capa
citors, an
d
trans
formers
.
Po
w
e
r
s
y
s
t
em r
e
ac
tive
po
w
e
r
o
p
t
imiza
t
i
on is i
m
pl
emented
ba
sed on th
e i
m
prove
d
CPSO p
r
op
o
s
ed i
n
thi
s
p
aper,
wh
ose
cod
e
ad
opts
real
numb
e
r
with continu
o
u
s a
nd
discrete
variable
s
mix
ed. The wh
ol
e step is d
e
scribed b
e
lo
w.
1)To in
put the initial para
m
eters, inclu
d
ing
po
we
r systems pa
ra
meters, cont
rol variable
s
,
and
con
s
trai
nt con
d
itions.
To set the
scale of
the
colony, and in
itialize the p
opulatio
n, wh
ich
inclu
d
e
s
the
displ
a
cement
Xi, the individual extr
emu
m
Pbest, a
n
d
the glob
al ex
tremum
Gbe
s
t of
each parti
cle,
and so o
n
.
2) For e
a
ch
particl
e in col
ony, tide current
and grid
loss cal
c
ulati
on are impl
e
m
ented,
according to (8) the fitness
value of whi
c
h is evaluate
d
, and Pbest,
and Gbe
s
t are update
d
.
3) To jud
ge
wheth
e
r the
mutation thre
shol
d co
nditi
ons a
r
e a
rriv
ed at, if the
con
d
ition
s
hold, then
mu
tation ope
rati
on is i
m
plem
ented a
c
cordi
ng to defin
e 1
.
Let Ex= Gb
est, En=
2Gb
e
st,
He
= En /10, t
hen a
c
cordin
g to define
2
mutation o
p
e
r
ation i
s
com
p
leted. Th
ose
parti
cle
s
whi
c
h
can n
o
t rea
c
h
mutation thre
shol
ds turn to the fouth step dire
ctly.
4) To perform evolution operato
r
for ea
ch
parti
cle. Let Ex= Pbest, En= 2Pbest, He= En
/10, accordin
g to define 2
a new pa
rticle j is
gene
ra
ted, and let i=j, thus evolu
t
ion operatio
n is
compl
e
ted.
5) If iteration times arrive
at the ma
ximum times, then output the Gbe
s
t, and the
optimizatio
n pro
c
e
ss e
n
d
s
. Otherwi
se t
u
rn to the 2
nd
step.
Evaluation Warning : The document was created with Spire.PDF for Python.
TEL
K
5.
R
exa
m
node
is ba
adju
s
maxi
m
27-2
8
The
u
foun
d
thres
the i
n
sho
w
of th
e
T
A
l
C
I
ratio
Tabl
e
K
OM
NIKA
Cl
o
R
esults a
nd
To verify
m
ple to illust
r
s, and 6
ge
n
lan
c
e no
de,
s
ting ste
p
-le
m
um tap
po
8
are tra
n
sf
o
u
pp
er an
d l
o
d
in Tab.
1. B
Table 1.
A
In the s
a
m
hold
N i
s
se
t
n
ertia we
ight
w
s t
he
co
mp
a
e
optimal val
T
able 2. Co
m
l
gorithm
PSO
C
PSO
CPSO
Und
e
r th
e
of all the tra
n
e
2, usi
n
g IC
Node
o
ud
Pa
r
t
ic
le
S
Discu
ssi
on
the effectiv
e
r
ate. IEEE3
0
n
erator no
d
e
the rest for
n
g
th of t
he
sit
i
o
n
s of
t
h
e
o
rme
r
br
an
c
h
o
wer limitati
o
enchm
ark p
o
A
ctive power
gen
e
m
e i
n
itial co
n
t
by 2, an
d t
h
is based
o
n
a
red results
o
ues after
50
-
m
pari
s
on of t
h
Power
loss(
p
0.0608
0.0591
0.0573
e
initial stat
e
n
sf
o
r
me
r
s
a
r
PSO metho
d
nu
m
b
e
r
1
—
2
5
8
11
13
I
S
S
warm
Algo
r
e
ne
ss of the
0
bus sy
st
e
m
e
s, wh
ere th
e
PV nodes,
1
capa
cito
r 1
e
two a
r
e fi
v
h
es, the sco
p
o
n of the re
a
o
wer i
s
set
a
Figure 1. I
E
limits of PV
e
ration of
P
V
n
ditio
n
s, i.e.
,
h
e maximu
m
n
(4
), ICPSO
o
f the th
ree
f
-
time
s opti
m
i
h
e
optimal r
e
p
u) Grid
e
s
,
the termi
n
r
e 1.0, the
s
y
d
, after the r
e
PG
Q
G
ma
x
/
——
0.
5
9
0.
8 0.
4
8
0.
5 0.
6
0.
2 0.
5
3
0.
2 0.
1
5
0.
2 0.
1
5
S
SN: 2302-4
r
ithm
Im
pro
v
prop
o
s
ed
m
m
, as
s
h
ow
n
e
node 1, 2,
1
0 and 24 is
0 i
s
0.1, t
h
v
e-po
sition.
p
e of the tra
n
a
ctive power
a
s SB = 100
M
E
EE 30 nod
e
nodes and
u
V
nodes a
n
d
,
the size of
m
iteration ti
m
is com
pare
d
f
or IEEE 30-
n
i
zation.
e
sul
t
s f
o
r dif
f
e
loss rate fal
ling
21.01
23.21
25.16
n
al voltage
s
y
s
t
em total
a
e
active po
w
e
/
pu
Q
G
mi
n
/
p
u
9
6
-
0
.
298
8
-
0
.
24
6
-0
.3
3
-
0
.
265
5
-
0
.
075
5
5
-
0
.
0775
046
v
em
ent and
A
m
ethod, we
t
a
n
in Figure
5, 8, 11, 13
reactive po
w
h
e on
e of t
h
Subci
r
cuits
i
n
sf
or
mer t
r
a
n
and voltage
M
V
A
.
e
bu
s sy
st
e
m
u
pp
e
r
and
lo
w
the balan
ce
the po
p
u
lati
o
m
es i
s
given
d
with CPS
O
n
ode syste
m
e
rent
algorit
h
(%)
Q
u
of all th
e g
e
a
ctive po
we
r
e
r optimi
z
ati
o
u
U
G
ma
x
/pu
1.
1
1.
1
1.
1
1.
1
1.
1
1.
1
A
pplication
…
a
ke bu
s I
E
E
E
1
,
contains
is the gene
r
w
er com
p
en
s
h
e nod
e 24
i
n
c
ludi
ng 6
-
9
n
sf
orm rat
i
o
s
l
e
vel of the
w
limits of
re
node
o
n
is set as
by 500, an
d
O
and PSO a
m
which are
t
h
h
m
s
in
bus I
E
u
alified rate of t
h
e
ne
rators an
d
grid loss is
0
o
n, the syste
U
G
mi
n
0.
9
0.
9
0.
9
0.
9
0.
9
0.
9
…
(Hon
gshe
n
E
30 sy
st
e
m
41
lines, 2
2
r
at
or nod
e
s,
s
at
ion node
s
i
s
0.02,
a
n
9
, 6-10,
4
-
1
2
s
are
±
8
×
1
generator
c
ac
tive powe
r
10
0, and
m
u
the adj
ust
m
lgorithm
s. T
a
h
e a
v
erage
v
E
E
E
30 sy
st
e
h
e voltage (%
)
90
100
100
d
the transf
o
0
.
0
771. See
n
m total pow
e
n
/pu
9
9
9
9
9
9
n
g S
u
)
473
as a
n
2
load
an
d 1
s
. The
n
d
t
he
2
, and
.25%.
c
an be
r
u
tation
m
ent o
f
a
bl
e 2
v
al
ue
s
e
m
o
rmer
-
n
from
e
r l
o
s
s
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 468 – 4
7
5
474
redu
ce
s to 0.
0573, an
d gri
d
loss rate le
ssen
s 25.
16
%. Clearly, this re
sult is b
e
tter than the
one
s
of the PSO a
nd CPSO.
Hence, ICPSO
algorith
m
is
a more effect
ive method. It can
acquire
th
e
global
optima
l
solutio
n
mo
re po
ssi
ble th
an PSO a
nd
CPSO, and
the voltage
a
m
plitude of
e
a
ch
node a
nd the
scope of ea
ch gene
rato
r reactive po
we
r
are n
o
t out of limits. The optimize
d
co
ntrol
variable
s
a
r
e
sho
w
n in Ta
b
l
e 3.
Table 3. Valu
es of co
ntrol
variable
s
afte
r optimization
Control variable
Nodes numbe
r
Voltage /pu
C
ontrol variable
Nodes numbe
r
Tap position
V
1
1
1.0733
T
1
4-12
6.0000
V
2
2
1.0703
T
2
6-9
3.0000
V
5
5
1.0409
T
3
6-10
1.0000
V
8
8
1.0490
T
4
27-28
1.0000
V
11
11
1.0666
Q
10
10
2.0000
V
13
13
1.0727
Q
24
24
3.0000
Figure 2. Conve
r
g
ence cu
rves
of PSO, CPSO and ICPSO
algorithm
s
Figure 2 sho
w
s the
co
nve
r
gen
ce
cu
rve
s
of
the thre
e algo
rithms.
Seen from
Figure 2,
there i
s
a fa
st falling befo
r
e
12 times i
n
CPSO
convergen
ce curve, but
hereafter
the cu
rve dro
p
s
very slowly, and the final conve
r
ge
nce effect is
not ideal. PSO curve ha
s a fast falling at
the
initial term, b
u
t at later stage, the curve
falls in
to sta
ndstill state due to
the early-maturin
g of the
particl
es. ICP
S
O curve de
scend
s very fast at t
he very start, and
then is followed by a buffer
stage b
e
twee
n 3 and 12 int
e
ration time
s,
hereafte
r dr
o
p
s sl
owly and
arrive
s at the optimal value
at aroun
d 23
times.
6. Conclusion
The ICPSO
proposed in this paper
fully
makes
use of
rando
mness and
stability
cha
r
a
c
teri
stics of the
clou
d dro
p
to re
alize
th
e co
mbination
of the global
a
nd lo
cal see
k
in
g
based on
sol
u
tion space transfo
rmatio
n
techn
o
logy,
a
nd al
so realize the cro
s
sover an
d mutati
on
operations using normal
cloud
operat
o
r. By applying in IEEE
30 bus
system, the simul
a
tio
n
results prove
that the improved
CPSO algorith
m
posse
sses d
r
am
atic improve
m
ent in see
k
i
ng-
optimization speed and preci
s
io
n, and comput
ation efficiency
and convergenc
e stability
are
better.
Referen
ces
[1]
Cha
ohu
a Da
i, W
e
iron
g Che
n
,
Yunfang Z
h
u
,
Xue
x
ia Z
h
a
n
g
. Reactive P
o
w
e
r Dis
patch
Consi
deri
n
g
Voltag
e Stabi
lit
y
w
i
th S
eeker
Optimizatio
n
Al
gorithm.
Electr
Power Syst Res
. 2009; 79(
10)
: 1462–
71.
[2]
AlRas
h
id
i MR,
El-Ha
w
a
r
y
ME.
A Survey
of Particle
S
w
arm Optimizatio
n
Applic
atio
ns in Electric Po
w
e
r
S
y
stems.
IEEE Trans Evol Com
p
ut
. 2009; 1
3
(
4): 913–
91
8.
[3]
Gomes JR, Saave
d
ra
OR.
Optimal React
i
ve Po
w
e
r Di
spatch Usin
g Evoluti
onar
y
Comp
utation:
Exte
nde
d Alg
o
r
ithms.
IEE Proc – Gener T
r
an
sm Distri
b
. 199
9; 146(6): 5
86-
592.
Pu/Grid Loss
Itera
tion
T
im
es
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Clou
d Particl
e
Swarm
Algorithm
Im
provem
ent and Application … (Hon
gsheng S
u
)
475
[4]
B Baran, J
Vall
ejos, R Ramos, U F
e
rnan
dez.
Multi-
obj
ective Re
a
c
tive Pow
e
r
Co
mp
ensati
o
n
.
IEEE/PES
T
r
a
n
smissio
n
an
d Distributi
on C
o
nferenc
e an
d Ex
p
o
siti
on. Atlan
t
a. 2001; 34
7-3
52.
[5]
F
u
rong L
i
, JD Pilgrim, C Da
b
eed
in, A Cheb
bo, RK
Aggar
w
a
l. Genetic Alg
o
rithms for Optimal Re
active
Po
w
e
r Compensation on
National Grid Sy
st
em.
IEEE Trans. Power Syst
.
200
5; 20(1): 49
3–5
00.
[6]
MS Osman, MA Abo-Si
nna
, AA Mousa.
A Soluti
on to
the Optimal
Po
w
e
r F
l
o
w
Using G
enetic
Algorit
hm.
Appl. Math. Com
put
. 2004; 155(
2): 391–
40
5.
[7]
H W
ang, H Su. Resaerch o
n
Capac
itors Optimi
zatio
n
Placem
ent in Di
stributio
n Net
w
ork Based o
n
Improved Ant
Colo
n
y
Alg
o
rith
m. Po
w
e
r S
y
st
em & Automati
on. 201
0; 32(6)
: 53-56.
[8]
W
enxia D
a
i, J
i
e W
u
. A Mod
i
fied Ge
netic
Algorit
hm
w
i
th
Anne
ali
ng Se
lectio
n for Re
active Po
w
e
r
Optimization.
Power System
Technology
. 200
1; 25(11): 3
3
-3
7
[9]
Mithun M. Bha
skar, S
y
d
u
lu
Mahes
w
a
ra
pu.
A H
y
br
id Gen
r
tic Algorithm
Appro
a
ch for
Optimal Po
w
e
r
Flo
w
.
T
E
LKOM
NIKA Indon
esi
an Jour
nal
of Electrical E
ngi
ne
erin
g
. 201
1; 9(2): 211-2
16.
[10]
Xu
e
x
i
a
Z
hang, W
e
iron
g Chen,
Chaoh
ua Dai. D
y
nam
ic Multi-
Group Self Adaptiv
e Differe
ntial Evol
uti
o
n
Algorit
hm for Reactive
Po
w
e
r
Optimization.
Int J Electr Power Energy Syst
. 2010; 32(
5): 351–
35
7.
[11]
Ar
y
a
LD, T
i
tare LS, Kothari
DP. Improved
Part
icle S
w
arm Optimization A
ppl
ied to R
eactive Po
w
e
r
Reserv
e Max
i
mization.
Int J
Electr Power Energy Syst.
20
10; 32(5); 3
68-
374.
[12]
Guang
w
e
i Z
h
a
ng, Rui Hu, Yu Liu, De
y
i
Li
. An Evolutio
nar
y
Algorit
hm Base
d on Clou
d
Model.
Chi
nes
e
Journ
a
l of Co
mputers
. 200
8; 3
1
(7): 108
2-1
0
9
1
.
[13]
L Z
hou, H Wang, W
W
ang.
Paralle
l Imple
m
entatio
n of
Classific
a
tio
n
Algorit
hms Based on Clo
ud
Computing Enviromrnt.
T
E
LKOMNIKA Indones
ian Jo
urn
a
l of
Electrical
Engine
eri
n
g
. 201
2; 10(5):
108
7-10
92.
[14]
De
yi L, Ch
eu
n
g
D, Shi XM
,
Ng V. Uncerta
i
nt
y
R
eas
oni
ng
Based o
n
Cl
oud Mo
dels i
n
Control
l
ers.
Co
mp
uters an
d Mathe
m
atics
w
i
th Applicati
o
ns
. 1998; 3
5
(3)
:
99–12
3.
[15]
Di K, De
y
i
L, D
e
ren L. Clo
ud
T
heor
y
an
d Its
Appl
icatio
ns in
Spatia
l Data Minin
g
Kno
w
l
e
dg
e Dscover
y
.
Journ
a
l of Ima
ge an
d Graph
i
c
s
. 1999; 4(1
1
)
:
930–9
35.
[16]
B Venkatesh, G Sadasivam,
Khan Ab
dul
lah
.
A Ne
w
Optim
a
l Reactiv
e
Po
w
e
r Sched
uli
n
g Method for
Loss Minim
i
zat
i
on an
d Volta
g
e
Stabilit
y Mar
g
in Ma
xim
i
zati
on Usin
g Succ
essive Multi-O
b
jectiv
e Fuzz
y
LP T
e
chniqu
e.
IEEE Trans Power Syst
. 2000
; 15(2): 844-8
5
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.