Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
10
,
No.
4
,
A
ugus
t
2020
,
pp. 343
1~34
40
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v10
i
4
.
pp3431
-
34
40
3431
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om/i
nd
ex
.ph
p/IJ
ECE
Solving
practical
economi
c load di
spatch p
ro
bl
em
usin
g crow sea
rch algo
rith
m
Sha
im
aa R. S
pea
El
e
ct
ri
ca
l
Eng
in
ee
ring
Depa
r
tment,
Fa
cul
t
y
of En
gine
er
ing, Meno
ufi
y
a
Univ
ersity
,
Eg
y
pt
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Sep
25
, 201
9
Re
vised
Jan
18
,
20
20
Accepte
d
Fe
b 2
, 2
020
The
pra
ct
i
ca
l
econom
ic
loa
d
dis
pat
ch
probl
em
is
a
non
-
conve
x
,
non
-
sm
ooth,
and
non
-
li
ne
ar
opti
m
iz
ation
pro
ble
m
due
to
includin
g
pra
ct
i
cal
conside
ra
ti
ons
such
as
val
ve
-
poi
nt
loa
ding
eff
ec
t
s
and
m
ult
ipl
e
fue
l
opti
ons
.
An
opt
imiza
t
ion
al
gori
thm
named
cro
w
sea
rch
algorithm
is
proposed
in
thi
s
pape
r
to
solve
t
he
pra
c
tical
non
-
conve
x
e
cono
m
ic
loa
d
disp
atch
proble
m
.
Thre
e
ca
ses
with
diffe
ren
t
ec
ono
m
ic
loa
d
dispatc
h
conf
igura
t
ions
are
studie
d
.
The
sim
ulation
result
s
and
st
at
isti
cal
an
aly
si
s
show
the
ef
fic
i
ency
o
f
the
proposed
c
row
sea
rch
al
g
orit
hm
.
Also,
t
he
sim
ula
t
ion
result
s
ar
e
compare
d
to
th
e
othe
r
rep
ort
e
d
al
gori
thms
.
The
compar
ison
of
result
s
conf
irms
the
hi
gh
-
qual
ity
solu
tions
and
the
eff
ec
t
ive
ness
of
th
e
proposed
al
gorit
hm
f
or
solving
the
non
-
conve
x
pra
ct
ical
e
c
onom
ic
loa
d
dispat
ch
problem
.
Ke
yw
or
d
s
:
Crow sea
rch al
gorithm
Eco
no
m
ic
load dis
patch
Mult
iple fue
l o
ptions
Valve
-
point e
ffec
ts
Copyright
©
202
0
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
:
Sh
ai
m
aa Ra
bah
S
pea,
Dep
a
rtm
ent o
f El
ect
rical
En
gi
neer
i
ng,
Faculty
of E
ngineerin
g, Me
no
uf
iy
a
Un
i
ver
sit
y,
Sh
e
bin
El
-
kom
, E
gypt.
Em
a
il
: shi_sp
e
a@ya
hoo.com
1.
INTROD
U
CTION
Eco
no
m
ic
load
disp
at
c
h
(EL
D)
is
a
n
essent
ia
l
op
tim
iz
at
io
n
ta
sk
in
t
he
powe
r
syst
em
.
It
rep
res
ent
s
a
basic
prob
le
m
in
the
po
w
er
syst
e
m
op
erati
on,
w
hich
obj
ect
s
to
ach
ie
ve
the
m
inim
u
m
cost
of
energy
requirem
ents
wh
il
e
sat
isfyi
ng
al
l
the
un
it
a
nd
syst
em
const
raints
[1
]
.
I
n
the
sim
plest
form
ulati
on
of
th
e
E
L
D
pro
blem
,
t
he
fu
el
cost
functi
on
of
the
ge
ne
rati
on
un
it
is
rep
rese
nted
by
a
qu
a
dr
at
ic
functi
on,
an
d
the
valv
e
po
i
nt
loa
ding
eff
ect
s
(VPL
)
are
i
gnor
e
d,
wh
ic
h
hav
e
the
ad
va
nta
ge
s
of
bei
ng
s
m
oo
th
an
d
c
onve
x.
These
a
dvanta
ges
inc
rease
t
he
nu
m
ber
of
op
tim
iz
at
ion
m
e
t
hods
t
hat
can
e
asi
ly
i
m
ple
m
e
nt
to
fi
nd
the
s
olu
ti
on
for
the
EL
D
pr
ob
le
m
.
In
prac
ti
cal
op
erati
ng
conditi
ons
of
the
po
wer
syst
e
m
,
m
any
therm
al
gen
erati
on
un
it
s
are
s
upplied
w
it
h
dif
fer
e
nt
s
ources
of
f
uel,
s
uch
as
nat
ur
al
gas,
oil,
an
d
c
oal.
It
is
neces
sary
to
fi
nd
th
e
m
os
t
econom
ic
al
fu
el
to
be
us
e
d
in
these
unit
s
[2
]
.
T
o
m
od
el
the
m
ulti
ple
fu
el
op
ti
on
s
(
MFO),
the
pie
cewise
qu
a
drat
ic
functi
on
is
us
e
d
f
or
the
representa
ti
on
of
f
uel
co
st
fu
nctio
n
[
3].
The
pract
ic
al
ELD
with
VPL
an
d
MFO
is
a
no
n
-
conve
x,
non
-
c
on
ti
nu
ous
,
a
nd
non
-
diff
e
ren
ti
able
opti
m
iz
a
tio
n
pr
ob
le
m
with
m
any
equ
al
i
ty
and
ineq
ualit
y con
s
trai
nts, wh
ic
h m
akes it ver
y
di
ff
ic
ult t
o fin
d
t
he op
ti
m
al
so
luti
on of t
his pr
ob
le
m
[
4].
Fo
r
it
s
im
po
rtance,
m
any
res
earche
rs
try
t
o
so
l
ve
the
EL
D
prob
le
m
us
ing
a
ver
it
y
of
conve
ntion
a
l
and
non
-
c
onve
ntion
al
m
et
ho
ds
.
T
he
co
nv
e
ntion
al
m
et
ho
ds
su
c
h
as
Quadr
at
ic
Progra
m
m
ing
[5
]
and
Linea
r
Pr
og
ram
m
ing
[6
]
oft
en
fail
to
obta
in
t
he
be
st
so
luti
ons
t
o
the
non
-
c
onvex
pro
blem
s
as
they
ass
ume
d
that
the
functi
ons
are
sm
oo
th
and
conve
x.
Als
o,
the
conve
rg
e
nc
e
of
these
m
e
thods
de
pends
on
the
init
ia
l
po
ints,
and
they
a
re
easy
to
c
onve
rg
e
into
the
local
op
ti
m
al
s
olu
ti
on.
Th
us
m
any
of
the
c
onve
ntion
al
m
et
hods
are
not
ef
fici
ent
to
fi
nd
t
he
so
luti
on
of
t
he
EL
D
prob
l
e
m
,
especial
ly
wh
e
n
t
he
pra
ct
ic
al
con
diti
ons
ar
e
consi
der
e
d.
To
ov
e
rco
m
e
the
lim
it
ation
s
of
the
cl
assic
al
m
et
ho
ds,
va
rio
us
E
vo
l
ution
a
r
y
Algorithm
s
(EA
s
)
hav
e
been
i
m
ple
m
ented
t
o
so
lve
the
ELD
pro
blem
su
ch
as
Soci
al
Sp
ider
Algorithm
(S
SA
)
[4
]
,
Gen
et
ic
Algo
rithm
(G
A
)
[7
]
,
Chaotic
Ba
t
A
lgorit
hm
(CBA)
[8
]
,
Ele
pha
nt
Her
di
ng
Op
ti
m
iz
at
ion
(EHO)
[
9],
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
4
,
A
ugus
t
2020
:
3431
-
3440
3
432
Ba
cktrack
i
ng
Searc
h
Al
gorithm
(BSA
)
[
10]
,
Moth
Flam
e
Algorithm
(MFA)
[
11]
,
et
c.
Most
of
these
al
gorithm
s
hav
e
fast
converg
ence
cha
racteri
sti
cs
and
high
pr
eci
sio
n.
S
o
they
can
deal
m
or
e
eff
ect
ivel
y
and
rob
us
tl
y
with
pr
act
ic
al
and
l
arg
e
-
scal
e
pro
blem
s.
In
this
pap
e
r,
a
Crow
Searc
h
Algorit
hm
(
CSA)
is
pr
op
os
ed
to
so
lve
t
he
no
n
-
c
onve
x
pract
ic
al
ELD
pro
bl
e
m
con
side
rin
g
V
PL
an
d
M
FO
.
T
he
pro
posed
CS
A
is
te
ste
d
on
10
-
un
it
te
st
syst
e
m
,
la
rg
e
scal
e
te
st
sys
tems
with
30,
60,
an
d
100
un
it
s,
a
nd
ve
ry
la
rg
e
-
sc
al
e
te
st
sys
tems
with
500,
1500,
2000,
a
nd
2500
unit
s
.
The
sim
ula
ti
on
res
ults
are
com
par
ed
to
ot
her
rele
va
nt
r
eported
al
gorithm
s.
The
oth
e
r
sect
ion
s
of
the
pa
pe
r
are
a
rr
a
nge
d
as
f
ollows:
T
he
m
at
he
m
atic
al
fo
rm
ulati
on
of
non
-
c
onve
x
ELD
pro
blem
is
pr
e
sented
i
n
sect
ion
2.
The
des
cripti
on
of
the
pro
po
se
d
CS
A
is
gi
ven
i
n
sect
ion
3.
Sect
ion
4
descr
i
bes
ho
w
the
CSA
is
app
li
ed
t
o
the
ELD
pr
ob
le
m
.
The
sim
ulati
on
resu
lt
s,
sta
ti
sti
cal
analy
si
s,
an
d
com
par
ison res
ults ar
e
sho
wn
in secti
on
5.
Se
ct
ion
6
s
hows
the c
on
cl
us
io
n of t
he pa
per
.
2.
PROBLE
M
F
ORMUL
ATI
ON
The
no
n
-
sm
oo
t
h
qua
dr
at
ic
cos
t
fu
nctio
n
is
m
or
e
acc
ur
at
e
in
the
represe
ntati
on
of
the
EL
D
pro
blem
.
VP
L
an
d piece
wise
qu
a
dr
at
ic
functi
ons
du
e
to
MF
O
a
re e
xa
m
ples o
f
thi
s t
ype of c
os
t
fun
ct
ion
s.
2.1. Ob
jecti
ve
f
unc
tio
n
ELD
with
val
ve
-
point
loa
di
ng
ef
fects:
Wh
en
the
ste
am
adm
issi
on
va
lves
are
ope
ne
d
to
co
ntr
ol
the
outp
ut
power
a
nd
to
obt
ai
n
hi
gh
e
r
power
le
vels
f
rom
the
ge
ner
at
ion
unit
s,
a
s
ha
rp
i
ncr
ease
in
thr
ottl
in
g
lo
sses
occ
urrs.
This
c
auses
ri
pp
le
s
in
t
he
fuel
-
cost
c
urve
[
7].
As
the
val
ve
is
pro
gressi
ve
ly
li
fted,
thes
e
losse
s
decr
ease
unti
l
the
val
ve
is
com
plete
ly
op
en
.
T
his
is
known
as
V
P
L,
w
hich
can
be
m
at
he
m
atical
l
y
m
od
el
le
d
as fo
ll
ow
s:
(
)
=
+
+
2
+
|
sin
(
ƞ
(
−
)
)
|
(1)
ELD
with
m
ulti
ple
fu
el
s:
Pra
ct
ic
al
ly
,
m
u
ltiple
so
urces
of
fu
el
can
be
us
e
d
in
the
therm
al
po
we
r
sta
ti
on
s
to
sup
ply
the
gen
e
ra
ti
on
un
it
s.
I
n
this
case,
t
he
piecew
ise
qua
dr
at
ic
c
os
t
f
un
ct
ion
will
be
m
or
e
su
it
able
in
the
represe
ntati
on
of
fu
el
c
os
t
for
dif
fer
e
nt
f
uel
ty
pes.
He
nce,
the
obj
ect
ive
of
the
ELD
prob
le
m
with
piece
wise
fu
el
c
os
t
f
un
ct
ion
is
to f
in
d
t
he
m
ini
m
u
m
total
fu
el
cost
a
m
on
g
the
a
vaila
ble
f
uels
of
ea
c
h
unit
wh
il
e sati
sfyin
g
the
syst
em
c
on
st
raints
[2
-
4].
T
his can
be
m
at
hem
atical
l
y fo
rm
ulate
d
as
f
ollows:
(
)
=
{
1
+
1
+
1
2
≤
≤
1
1
2
+
2
+
2
2
1
≤
≤
2
2
.
.
.
+
+
2
−
1
≤
≤
(2
)
wh
e
re
,
α
iL
,
β
iL
and γ
iL
a
re th
e
cost c
oeffici
ents of the
i
-
th
g
e
ner
at
or
f
or
t
he fuel ty
pe
L
.
ELD
with
m
ulti
ple
fu
el
s
an
d
valve
-
point
loadi
ng
e
ff
e
ct
s:
The
fuel
cost
functi
on
wh
e
n
VPL
an
d
MFO a
re c
onsidere
d
ca
n be
re
pr
ese
nted
as
fol
lows
:
(
)
=
{
1
+
1
+
1
2
+
|
1
s
in
(
1
(
−
)
)
|
≤
≤
1
1
2
+
2
+
2
2
+
|
2
s
in
(
2
(
−
)
)
|
1
≤
≤
2
2
.
.
.
+
+
2
+
|
si
n
(
(
−
)
)
|
−
1
≤
≤
(3
)
2.2. C
on
s
tr
aint
s
Power
balance
con
st
raint:
Th
e
total
power
gen
e
rati
on
m
us
t
sat
isfy
the
t
otal
load
dem
a
nd
(
P
D
)
a
nd
the tra
ns
m
issi
o
n powe
r
los
ses
(
Ploss
) [4
]
.
H
e
nce,
∑
=
+
=
1
(4
)
Gen
e
rati
on
li
m
it
s
con
strai
nt:
The
real
outp
ut
power
f
ro
m
each
gen
e
rato
r
m
us
t
be
betwe
en
it
s
m
ini
m
u
m
and
m
axi
m
u
m
lim
i
ts as f
ollo
ws [1]
:
≤
≤
,
=
1
,
…
…
.
.
,
(5)
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
So
lv
in
g
pract
ic
al econ
om
ic
l
oad dis
patc
h pr
ob
le
m usin
g
cr
ow se
ar
c
h alg
ori
thm
...
(
Shai
maa R.
Spea
)
3433
3.
C
R
OW
SEA
R
CH ALG
ORIT
HM
Crow
sea
rch
a
lgorit
hm
(CSA)
is
a
m
et
a
-
he
ur
ist
ic
opti
m
izati
on
al
go
rith
m
pr
esented
by
Ask
a
rza
deh
in
20
16
[
12]
.
The
m
ai
n
idea
of
the
CS
A
i
s
obta
ine
d
fro
m
no
ti
ci
ng
the
so
ci
al
be
hav
i
or
of
c
rows
,
wh
ic
h
consi
der
the
m
os
t
intel
li
gen
t
bir
ds
.
Cr
ow
s
li
ve
in
the
f
or
m
of
floc
ks.
The
y
char
act
erize
d
by
ha
ving
a
good
m
e
m
or
y
[13]. Crows
a
re
t
hieves.
Th
ey
watc
h
oth
e
r
bir
ds
,
inclu
ding
t
he
oth
er
cr
ow
m
e
m
ber
s
on
the f
lo
ck,
a
nd
ob
s
er
ve
w
here
they
hid
e
thei
r
f
ood.
O
nce
the
ot
her
bir
ds
l
eave,
they
ste
al
their
foo
d.
The
cr
ows
use
their
intel
li
gen
ce
to
hid
e
t
heir
e
xcess
foo
d
in
a
hid
e
out
spot
an
d
re
store
this
f
ood
w
hen
t
hey
nee
d
[
14]
.
It
is
diff
ic
ult
to
fin
d
the
cr
ow
st
or
e
d
foo
d.
If
a
c
row
dis
cov
e
rs
a
nothe
r
one
is
goin
g
after
it
,
it
will
tr
y
to
deceive
that
c
r
ow
an
d
will
go
to
a
no
t
her
posit
ion
[
12
]
.
T
his
intel
li
gen
t
m
ann
er
of
t
he
cr
ow
s
is
sim
i
la
r
to
the
op
ti
m
iz
ati
on
process,
and
CS
A
at
te
m
pts
to
sim
ul
at
e
that
beh
a
vi
or
t
o
fin
d
th
e
opti
m
a
l
so
l
utions
to
the
optim
iz
at
io
n
pro
blem
s
[1
2]
.
If
the
re
is
a
s
olu
ti
on
s
pace
with
dim
ensio
n
d
has
a
c
row
fo
l
k
of
n
c
r
ow
s
,
then
the posit
io
n
X
of cr
ow
i
at
it
erati
on
t
can
b
e
expresse
d by t
he vect
or:
,
=
[
1
,
,
2
,
,
…
…
,
,
]
(6
)
w
he
re
:
i
=1:
n
,
t
=1:
t
max
, and
t
max
is t
he
m
axi
m
u
m
n
u
m
ber
of it
erati
ons.
Each
cr
ow
ha
s
a
m
e
m
or
y
,
in
wh
ic
h
it
st
or
es
th
e
best
po
sit
io
n
of
it
s
sto
rin
g
f
ood
so
urce.
The
vect
or
X
c
on
ta
in
s
the
ra
ndom
init
ia
l
po
s
it
ion
s
of
the
cr
ow
s
.
The
se
po
sit
ion
s
are
upda
te
d
at
each
it
erati
on,
and
this
pr
oces
s
is
rep
eat
ed
unti
l
the
stop
pi
ng
crit
erio
n
is
m
et
.
To
update
the
po
sit
io
ns
of
the
cr
ows,
there
ar
e
two
ca
ses
[12,
14
]
:
Ca
se
1
:
C
row
j
does
not
recogn
i
ze
that
c
row
i
is
goin
g
af
te
r
it
;
he
nce,
cr
ow
i
will
get
c
lose
the
sto
rin
g
plac
e
of cr
ow
j
. In
th
is st
at
e, the p
osi
ti
on
of c
row
i
will
b
e
update
d
as
foll
ows:
,
+
1
=
,
+
×
,
×
(
,
−
,
)
(7
)
w
he
re
:
ra
i
is a
r
a
ndom n
um
ber
w
it
h u
nif
or
m
d
ist
ributi
on
a
nd it
s v
al
ue betwee
n 0 a
nd 1.
fl
i,t
is t
he
fligh
t
le
ng
th
of c
row
i
at
it
erati
on
t
.
fl
has
a
n
ef
fect
on
the
capa
bili
ty
of
th
e
searc
h
[
12]
.
Adjusti
ng
the
value
of
fl
will
help
i
n
the
co
nver
ge
nc
e
of
the searc
h
al
go
rithm
[
13
]
.
Ca
se
2
:
Crow
j
recog
nizes
tha
t
crow
i
is
go
i
ng
a
fter
it
.
S
o,
it
will
m
ov
e
to
ano
t
her
posit
ion
t
o
decei
ve
c
row
i
and to
save
it
s f
oo
d.
The
s
umm
ary
of the t
w
o ca
se
s is as
fo
ll
ows:
,
+
1
=
{
,
+
×
,
×
(
,
−
,
)
≥
,
ℎ
(8
)
wh
e
re
r
j
is
the
unifo
rm
distribu
te
d
r
an
dom
nu
m
ber
i
n
the
r
an
ge
of
[
0,
1],
an
d
AP
is
t
he
a
war
e
ness
factor.
The
value
of
AP
co
ntr
ols
th
e
intensific
at
io
n
an
d
div
e
rsif
ic
at
ion
of
the
op
ti
m
iz
ation
process
.
Dec
re
asi
ng
the
val
ue
of
AP
will
increa
se
the
c
han
ce
of f
in
ding
the
sto
r
ing
food
s
ourc
es
by
the
cr
ow
s.
T
his
w
ould
a
m
pl
ify
the
inte
ns
ific
at
ion
of
the
al
go
rithm
[1
2].
H
oweve
r,
inc
reas
ing
t
he
value
of
AP
m
ay
m
a
ke
t
he
c
rows
s
earch
the
sp
ace
rand
om
l
y,
wh
ic
h
de
creasin
g
their
chan
ce
t
o
fin
d
the
stori
ng
food
s
ources
.
Th
is
le
ads
to
am
p
li
fyi
ng
the d
i
ver
si
ficat
ion
of the
alg
or
it
h
m
[
13]
.
Pse
udo co
de o
f
the
CSA ca
n b
e
de
scribe
d
as
sho
wn in
belo
w
.
B
egin
D
ef
ine
n,
fl
,
A
P
an
d
t
max
D
ef
ine
ob
je
ct
iv
e fun
ct
io
n,
decisi
on va
ri
ab
le
s
and
c
onstra
i
nts
I
niti
alize r
andomly t
he
posit
ion
s
o
f
ft
oc
k
of
n
cr
ows i
n
the
searc
h
s
pac
e
I
niti
alize
the
memor
y
of ea
c
h crow
(
init
ial m
emo
r
y=init
ial
po
sit
io
n
)
E
valu
ate
the
posit
ion
s
of t
he c
ro
ws
(
fi
tness
)
S
et
the it
er
atio
n
c
ounter t=
1
M
ain
lo
op
:
W
hil
e
(
t
<
t
max
)
F
or
i
=1
:
n
R
ando
mly
ch
oose
one
of the c
ro
ws t
o
foll
ow
(
for exam
ple
ch
oos
e j
)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
4
,
A
ugus
t
2020
:
3431
-
3440
3434
I
f
r
j
>
AP
jt
X
i,
t
+
1
=
X
i,
t
+
ra
i
×
fl
i,t
×
(
m
j,t
–
X
i,
t
)
E
lse
X
i,
t
+
1
=
a ra
ndom
po
sit
io
n of s
earc
h
s
p
ac
e
E
nd
if
E
nd
for
C
heck
the
feasi
bil
it
y o
f new
posit
ion
s
E
va
lu
ate
the
new
posit
ion
of t
he
cr
ows
U
pd
ate
the
me
mo
ry
o
f
cro
ws
E
nd
whi
le
F
ind
the
opti
omal s
ol
ution
E
nd
4.
APPLI
CA
TI
ON OF
CSA
TO EL
D
P
ROB
LE
M
In
this
pap
e
r,
t
he
m
ai
n
ste
ps
of
the
pro
pose
d
CSA
im
ple
m
entat
ion
to
s
ol
ve
the
EL
D
pr
ob
le
m
can
be
exp
la
ine
d as f
ol
l
ow
s:
Step
1:
De
fin
e
the
al
go
rit
hm
par
am
e
te
rs
includi
ng
n
,
t
max
,
fl
,
and
AP
,
and
def
i
ne
the
syst
e
m
const
raints
includi
ng up
pe
r
a
nd lo
wer
val
ues of p
ower
generati
on
unit
s and
powe
r bala
nce c
onstrai
nt
.
Step
2:
I
niti
al
i
zat
ion
of
the
posit
ion
a
nd
m
em
or
y
of
the
cr
ow
s:
Gen
e
rate
rando
m
ly
the
init
ia
l
po
pula
ti
on
of
crow
fo
l
k posi
ti
on
s
in
t
he
sea
r
ch
s
pace
us
in
g (9) as
foll
ows:
,
=
(
)
+
(
(
)
−
(
)
)
×
i
=1:n, k
=
1:d
(9)
wh
e
re
r
and
is
a
unif
or
m
ly
di
stribu
te
d
rand
om
n
um
ber
be
tween
0
a
nd
1.
,
is
a
m
at
rix
with
dim
ension
s
×
.
The
posit
ion
of
each
c
r
ow
ob
ta
i
ned
by
(
9)
re
pr
ese
nts
a
sugg
e
ste
d
s
olu
ti
on
to
t
he
ELD
pro
ble
m
.
The n
um
ber
of
contr
ol v
a
riabl
es
d
eq
uals the
nu
m
ber
of com
m
itted g
e
ner
at
i
on unit
s (
d=N
G
).
Nex
t,
ge
ner
at
e
the
crow
init
ia
l
m
e
m
or
y.
In
this
work,
it
is
su
ppose
d
th
at
the
init
ia
l
m
e
m
or
y
of
the cr
ows is t
he
sam
e as their initi
al
p
os
it
io
ns.
Step
3:
Eval
ua
te
the
obj
ect
iv
e
functi
on
a
nd
cal
culat
e
the
fitness
val
ue
f
or
eac
h
cr
ow
:
Ca
lc
ulate
the
fitness
value by s
ubsti
tuti
ng
t
he p
os
it
ion
s
into
the
fuel
co
st
obj
ect
iv
e f
un
ct
io
n, w
hi
ch
is
represe
nted by:
1.
E
quat
ion
(
1
)
w
hen VPL i
s c
onside
red o
r,
(
1)
2.
E
quat
ion
(
2
)
w
hen MFO
is c
onside
red o
r,
(
2)
3.
E
quat
ion
(
3
)
w
hen MFO
a
nd
VP
L
are c
onsi
der
e
d.
(
3)
Step
4:
Ge
ner
a
te
the
ne
w
posit
ion
s
of
cr
ows:
Fin
d
the
ne
w
po
sit
io
ns
of
th
e
crows
i
n
the
d
-
dim
ension
al
searc
h
sp
ace
as
fo
ll
ows:
I
f
cr
ow
i
l
ooks
for
a
new
posit
ion
,
it
wi
ll
ran
dom
ly
ch
oo
s
e
one
of
t
he
crows
j
a
nd
go
a
fter
it
to
disco
ver
the p
os
i
ti
on
of
it
s
hid
de
n
f
ood
sou
rce
s
(
m
j
)
.
The
ne
w
posit
ion
o
f
c
row
i
will
be
fou
nd
acco
rd
i
ng to
(8).
Step
5:
Che
cki
ng
t
he
feasi
bili
ty
of
ne
w
posit
ion
s:
Chec
k
th
e
feasibil
it
y
of
the
ne
w
posit
ion
of
eac
h
c
row,
a
nd
update
the
pos
it
ion
base
d
on
it
.
If
the
new
po
sit
io
n
is
fea
sible,
the
posit
ion
is
update
d,
and
if
no
t
,
the cr
ow r
em
a
ins i
n
it
s curre
nt
p
osi
ti
on
a
nd
do
e
s
no
t m
ov
e
to the ne
w posi
ti
on
fou
nd.
Step
6:
E
valua
te
the
obj
ect
iv
e
f
un
ct
io
n
of
new
posit
io
ns
:
Eval
uate
the
new
posit
io
ns
,
an
d
obta
in
th
e
ne
w
fitness
values
a
s explai
ne
d
in
ste
p 3.
Step
7: up
date
m
e
m
or
y:
U
pda
te
the cro
w
s m
e
m
or
y as
fo
ll
ows
[12]:
,
=
{
,
+
1
(
,
+
1
)
ℎ
(
,
)
,
ℎ
(10)
Eq
uation
(10
)
sta
te
s
that
if
the
fitness
of
new
posit
ions
is
bette
r
tha
n
the
f
it
ne
ss
of
m
e
m
or
y
posit
ion
s,
the
m
e
m
or
y i
s u
pdat
ed
.
Step
8:
En
d
th
e
al
go
rithm
if
the
stoppin
g
cr
it
erion
is
m
et
:
If
the
m
axi
m
um
nu
m
ber
of
it
erati
on
s
is
rea
ched,
En
d
the
alg
or
it
hm
.
Step
9:
Fin
d
t
he
op
ti
m
al
so
luti
on
:
Fin
d
th
e
optim
al
so
lu
ti
on
wh
ic
h
inc
lud
es
the
opti
m
al
ou
tp
ut
po
wer
of
gen
e
rati
on
un
it
s and its c
orres
pondin
g op
ti
m
al
v
al
ue o
f
tota
l fu
el
c
os
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
So
lv
in
g
pract
ic
al econ
om
ic
l
oad dis
patc
h pr
ob
le
m usin
g
cr
ow se
ar
c
h alg
ori
thm
...
(
Shai
maa R.
Spea
)
3435
5.
SIMULATI
O
N RESULTS
AND DIS
C
USSION
The
pro
po
se
d
CSA
is
im
plem
ented
in
M
ATL
AB
7.1
0.0
e
nv
ir
onm
ent.
The
pr
ogram
s
are
run
on
a
perso
nal
c
om
pu
te
r
with
a
n
I
ntel
Co
re
I5,
2.2
G
Hz
proc
essor,
4
GB
R
AM,
a
nd
the
W
i
ndows
8.1
op
e
rati
ng
syst
e
m
.
Du
e
to
the
ra
ndom
natur
e
of
CS
A,
sever
al
tria
ls
with
dif
fer
e
nt
init
ia
l
popu
la
ti
on
s
ar
e
ca
rr
ie
d
out
t
o
ob
ta
in
a
use
f
ul
concl
us
io
n
of
the
pe
rfor
m
ance
of
t
he
al
gorit
hm
and
to
c
hoos
e
the
best
va
lues
of
the
pro
pos
e
d
CSA
im
po
rtan
t
par
am
et
ers,
wh
ic
h
incl
ud
e
n
,
fl
,
a
nd
AP
.
To
optim
iz
e
these
pa
ram
et
ers,
se
ver
al
e
xp
e
r
i
m
ents
are ru
n by va
ry
ing
t
heir val
ue
s as foll
ows:
AP
is
cha
nged
from
0
to
1
with
a
ste
p
0.
05,
fl
is
cha
nged
f
ro
m
0
to
5
with
a
ste
p
0.1,
a
nd
n
i
s
change
d
from
50
to
25
0
with
ste
p
5.
T
he
value
of
one
par
am
et
er
is
c
hange
d
i
n
it
s
range
w
hile
th
e
ot
her
par
am
et
ers
are
fixe
d.
F
or
ea
ch
com
bin
at
io
n,
the
EL
D
pr
ob
le
m
is
so
lve
d,
a
nd
the
sta
t
ist
ic
al
ind
ic
es
of
t
he
obj
ect
ive
f
un
ct
ion
a
re
cal
culat
ed.
T
he
best
va
lues
of
the
pa
r
a
m
et
ers
wh
ic
h
gav
e
the
m
inim
u
m
cost
are
cho
s
en
as
the optim
al
set
ti
ng
s o
f
co
nt
ro
l par
am
et
ers.
It
is fou
nd
that
the
m
os
t
su
it
a
ble
val
ues
f
or
fl
and
AP
f
or
ca
ses
1
and 2 are
2.0 a
nd 0.1,
resp
ect
i
vely
, and f
or
c
ase 3
are 3.
0
a
nd 0.1.
5.1. C
as
e
1: E
LD wi
th
valve
-
po
in
t
l
oading
eff
ec
ts
In
this
case
,
t
he
perf
or
m
ance
of
the
propose
d
CS
A
i
n
so
lvi
ng
no
n
-
c
onve
x
E
LD
w
it
h
VPL
i
s
discusse
d.
The
10
-
unit
te
st
s
yst
e
m
[1
5]
is
adopted
f
or
this
stud
y.
The
te
st
syst
e
m
con
sist
s
of
te
n
generati
ng
un
it
s
with
loa
d
dem
and
20
00
M
W
.
Ta
ble
1
s
hows
t
he
syst
em
data.
The
po
wer
l
os
ses
a
re
neg
le
ct
e
d.
n
an
d
t
max
are
sel
ect
ed
to
be
60
a
nd
10000,
resp
e
ct
ively
.
The
tot
al
exec
ution
tim
e
of
CS
A
is
36.
786
sec,
a
nd
the
exec
ution
t
i
m
e
per
gen
e
ra
ti
o
n
is
0.0
03
s
ec.
The
power
disp
at
ch
res
ults
are
li
ste
d
in
Table
2
al
on
g
with
the
m
in,
m
ean,
an
d
m
ax
values
of
fu
el
c
ost
.
From
this
tab
le
,
it
is
ob
s
er
ved
t
hat
the
sy
stem
con
strai
nt
s
are
sat
isfie
d
s
ucce
ssfu
ll
y.
T
he
optim
al
value
of
the
c
os
t
obta
ined
by
th
e
p
rop
os
ed
C
SA
is
10
617.0
$/hr.
T
he
eff
ect
i
veness
of
the
pro
pose
d
CSA
is
com
par
ed
with PS
O
[
15]
,
MSC
O
[16],
an
d
PH
O
A
[
17
]
,
as
giv
e
n
in
Table
2.
It
is
cl
ear
that
the
propose
d
CSA
outpe
rfor
m
s
these
m
et
ho
ds
as
it
giv
es
the
best
values
of
th
e
m
in
and the m
ean c
o
st c
om
par
ed
t
o
the
o
t
her m
eth
ods.
Table
1.
Lim
it
s
of
gen
e
rati
on
un
it
s a
nd c
os
t
coeffic
ie
nts
for
10
-
unit
test
syst
e
m
Un
it
Min
Mw
Max
MW
α
$
/MW
2
β
$
/MW
γ
$
ρ
$
η
MW
-
1
U1
10
55
0
.12
9
5
1
4
0
.54
0
7
1
0
0
0
.4
0
3
33
0
.01
7
4
U2
20
80
0
.10
9
0
8
3
9
.58
0
4
9
5
0
.606
25
0
.01
78
U3
47
120
0
.12
5
1
1
3
6
.51
0
4
9
0
0
.705
32
0
.01
6
2
U4
20
130
0
.12
1
1
1
3
9
.51
0
4
8
0
0
.705
30
0
.01
6
8
U5
50
160
0
.15
2
4
7
3
8
.53
9
7
5
6
.799
30
0
.01
4
8
U6
70
240
0
.10
5
8
7
4
6
.15
9
2
4
5
1
.325
20
0
.01
6
3
U7
60
300
0
.03
5
4
6
3
8
.30
5
5
1
2
4
3
.5
3
1
20
0
.01
5
2
U8
70
340
0
.02
8
0
3
4
0
.39
6
5
1
0
4
9
.9
9
8
30
0
.01
2
8
U9
135
470
0
.02
1
1
1
3
6
.32
7
8
1
6
5
8
.5
6
9
60
0
.01
3
6
U1
0
150
470
0
.01
7
9
9
3
8
.27
0
4
1
3
5
6
.6
5
9
40
0
.01
4
1
Table
2.
Sim
ul
at
ion
resu
lt
s
for 10
-
unit
test
s
yst
e
m
Prop
o
sed
CSA
PSO [
1
5
]
MSCO [
1
6
]
PHOA [
1
7
]
P
1
5
5
.00
0
0
5
3
.10
0
0
5
5
.00
0
0
5
5
.00
0
0
P
2
8
0
.00
00
7
9
.20
0
0
8
0
.00
0
0
8
0
.00
0
0
P
3
8
9
.08
1
8
1
1
2
.000
9
1
.40
6
7
9
8
.27
9
2
P
4
8
0
.19
5
7
1
2
1
.000
7
3
.86
5
4
7
3
.29
4
3
P
5
6
6
.35
0
0
9
8
.80
0
0
7
0
.57
0
0
7
0
.22
7
8
P
6
7
0
.00
0
0
1
0
0
.000
7
0
.00
0
0
7
2
.70
2
5
P
7
2
9
0
.6553
2
9
9
.000
2
8
2
.6504
2
7
0
.4959
P
8
3
2
8
.7171
3
2
0
.000
3
4
0
.0000
3
4
0
.0000
P
9
470
.0000
4
6
7
.000
4
7
0
.0000
4
7
0
.0000
P
10
4
7
0
.0000
3
5
6
.000
4
6
6
.5075
4
7
0
.0000
Min. cos
t
1
0
6
1
7
.0
1
0
7
6
2
0
1
0
6
1
9
.8
1
0
6
2
1
.0
Mean cos
t
1
0
6
1
8
.0
-
1
0
6
3
2
.0
1
0
6
2
1
.0
Max. cos
t
1
0
7
9
6
.0
-
1
0
6
4
5
.0
1
0
6
2
1
.0
5.2. C
as
e
2: P
ie
cew
ise
qu
adratic
fu
el
c
os
t
The
e
ff
ect
ive
ne
ss
of
p
r
opos
e
d
CS
A
for
s
olvi
ng
t
he
EL
D
pr
ob
le
m
with
M
FO
is
stu
died
usi
ng
t
he
10
-
un
it
te
st
syst
e
m
,
la
rg
e
-
scal
e
te
st
syst
e
m
s
[3
]
,
and
ver
y
la
r
ge
-
scal
e
te
st
syst
e
m
s
[3
]
.
The
la
rg
e
-
scal
e
and
ver
y
la
rg
e
-
scal
e
te
st
syst
e
m
s
are
fo
rm
ed
base
d
on
the
10
-
un
it
te
st
syst
e
m
by
duplica
ti
ng
it
to
obta
in
th
e
re
qu
i
r
e
d
syst
e
m
siz
e.
The
first
te
st
sy
stem
has
10
unit
s.
Eac
h
unit
can
be
sup
plied
by
tw
o
or
three
dif
fere
nt
fu
el
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
4
,
A
ugus
t
2020
:
3431
-
3440
3436
Fo
r
c
om
par
iso
n
pur
po
ses
,
t
he
pro
blem
is
so
lve
d
for
loa
d
dem
and
s
2400
M
W,
2500
M
W
,
2600
M
W,
a
nd
2700
M
W.
T
he
powe
r
los
ses
are
neg
le
ct
ed
for
al
l
loa
d
de
m
and
s.
t
max
is
sel
ect
ed
to
be
10000.
T
he
disp
at
ch,
the
sta
ti
sti
cal
r
esults,
an
d
the
total
execu
ti
on
tim
e
are
li
st
ed
in
Ta
ble
3.
The
opti
m
al
fuel
costs
ob
ta
i
ne
d
by
the
pro
posed
CSA
are
481.722
6
$/
hr
,
52
6.238
8
$/
hr
,
574.3
808
$/h
r
,
and
623.8
092
$/
hr
f
or
the
load
s
2400
M
W,
25
00
M
W,
26
00
M
W
,
a
nd
2700
M
W
,
re
sp
ect
i
vely
.
Fig
ur
e 1
sh
ows
t
he
co
nver
ge
nce
cha
ra
ct
erist
ic
of
fu
el
c
os
t
at
diff
e
re
nt
load
dem
and
s.
T
he
good
pe
rfor
m
ance
of
pr
opos
e
d
CSA
i
n
the
gr
a
dual
decr
ea
se
of
th
e ob
j
ect
ive
f
un
ct
io
n u
ntil
r
e
achin
g
the
m
ini
m
u
m
v
al
ue
is
detect
ed fr
om
t
his f
i
gure.
Table
3.
Disp
at
ch
a
nd stat
ist
ic
al
r
esults
for 1
0
-
unit
test
syst
e
m
w
it
h
MF
O at
d
iffe
re
nt loa
d dem
and
s
Un
it
2
4
0
0
M
W
2
5
0
0
M
W
2
6
0
0
M
W
2
7
0
0
M
W
Fu
el
P
i
(
MW)
Fu
el
P
i
(
MW)
Fu
el
P
i
(
MW)
Fu
el
P
i
(
MW)
1
1
1
8
9
.7375
2
2
0
6
.5190
1
1
8
9
.4093
2
2
1
8
.1882
2
1
2
0
2
.3375
1
2
0
6
.4573
1
1
8
1
.3156
1
2
1
1
.6625
3
1
2
5
3
.8996
1
2
6
5
.7391
1
2
8
6
.8303
1
2
8
0
.7206
4
3
2
3
3
.0457
3
2
3
5
.9531
3
2
1
0
.8350
3
2
3
9
.5668
5
1
2
4
1
.8340
1
2
5
8
.0177
1
3
1
4
.8870
1
2
7
8
.4753
6
3
2
3
3
.0445
3
2
3
5
.9531
3
2
2
9
.7923
3
2
3
9
.5835
7
1
2
5
3
.2750
1
2
6
8
.8635
1
2
8
9
.7424
1
2
8
8
.6982
8
3
2
3
3
.0436
3
2
3
5
.9531
3
2
1
9
.1520
3
2
3
9
.5767
9
1
3
2
0
.3783
1
3
3
1
.4877
1
3
5
2
.8595
3
4
2
8
.4968
10
1
2
3
9
.4043
1
2
5
5
.0562
1
3
2
5
.1766
1
2
7
5
.0314
TPG* (
M
W
)
2400
2500
2600
2700
Min. cos
t
4
8
1
.7226
5
2
6
.2388
5
7
4
.3808
6
2
3
.8092
Mean cos
t
4
8
1
.8068
5
2
6
.3180
5
7
4
.4136
6
2
3
.8650
Max. cos
t
5
1
5
.8100
5
8
5
.0387
5
9
9
.1933
6
7
9
.6398
Std
1
.34
7
4
1
.36
3
3
0
.75
3
5
1
.28
5
9
Ti
m
e
(
sec
.)
1
0
.71
4
1
0
.81
3
1
0
.55
2
1
1
.44
5
*
T
PG
:
t
o
t
al
p
o
w
er ge
n
era
t
i
o
n
Figure
1
.
Co
nverg
e
nce c
ha
rac
te
risti
cs o
f
t
he pr
opos
e
d
C
SA
for 10
-
un
it
test
syst
e
m
w
it
h M
FO
The
c
om
par
iso
n
betwee
n
t
he
op
ti
m
al
fu
el
co
st
obta
ined
by
the
pr
opos
e
d
C
SA
a
nd
the
ot
he
r
re
porte
d
al
gorithm
s
is
giv
e
n
in
Table
4.
For
the
load
dem
and
2400
M
W
,
i
t
is
no
ti
ced
that
the
pr
op
os
e
d
CSA
giv
e
s
bette
r
f
uel
c
os
t
than
ARC
GA
[7
]
,
H
NU
M
[18],
an
d
M
PSO
[19]
a
nd
it
al
m
os
t
ob
ta
ins
t
he
sam
e
value
of
f
uel
cost
com
par
ed
to
oth
er
m
et
ho
ds.
For
loa
d
dem
and
2500
M
W
,
the
pro
pose
d
CSA
obt
ai
ns
the
sam
e
op
ti
m
al
fu
el
c
os
t
as
A
HNN
[
20]
,
an
d
it
giv
es
bette
r
fu
el
c
os
t
tha
n
the
oth
e
r
al
go
r
it
h
m
s.
Fo
r
l
oa
d
dem
and
2700
M
W,
the
pro
posed
CSA
ob
ta
ins
the
sam
e
fu
el
cost
as
Q
P
-
AL
HN
[
3],
RC
G
A
[
21]
,
HRCG
A
[
21]
,
an
d
M
PSO
[
19
]
and
it
gi
ves
be
tt
er
f
uel
co
st
c
om
par
ed
t
o
t
he
ot
her
m
et
ho
ds.
It
sh
ould
be
m
entioned
that
H
NU
M
[
18]
di
d
no
t
sat
isfy t
he pow
er
balance c
on
strai
nt for al
l l
oad d
em
and
.
0
100
200
300
400
500
600
700
800
900
1000
480
500
520
540
560
580
600
620
640
660
680
I
t
e
r
a
t
i
o
n
s
Fu
e
l
C
o
s
t
(
$
/
h
r
)
2
7
0
0
M
W
2
6
0
0
M
W
2
5
0
0
M
W
2
4
0
0
M
W
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
So
lv
in
g
pract
ic
al econ
om
ic
l
oad dis
patc
h pr
ob
le
m usin
g
cr
ow se
ar
c
h alg
ori
thm
...
(
Shai
maa R.
Spea
)
3437
Table
4.
C
om
par
iso
n of t
he b
est
f
uel c
os
t
for
10
-
unit
test
s
yst
e
m
w
it
h
MFO
Metho
d
2
4
0
0
M
W
2
5
0
0
M
W
2
6
0
0
M
W
2
7
0
0
M
W
Prop
o
sed
CSA
4
8
1
.722
5
2
6
.230
5
7
4
.380
623
.809
QP
-
A
LHN
[
3
]
4
8
1
.723
5
2
6
.239
5
7
4
.381
6
2
3
.809
ARC
GA [
7
]
4
8
1
.743
5
2
6
.259
5
7
4
.405
6
2
3
.828
AHNN [2
0
]
4
8
1
.720
5
2
6
.230
5
7
4
.370
6
2
6
.240
HGA [2
1
]
-
5
2
6
.240
5
7
4
.380
6
2
6
.810
RC
GA [
2
1
]
4
8
1
.723
5
2
6
.239
5
7
4
.396
6
2
3
.809
HRC
GA [
2
1
]
4
8
1
.722
5
2
6
.238
5
7
4
.380
6
2
3
.8
09
AIS
[
2
3
]
-
5
2
6
.240
5
7
4
.380
6
2
3
.810
HNUM
[
1
8]
4
8
8
.500
5
2
6
.700
5
7
4
.030
6
2
5
.180
MPSO
[
1
9
]
4
8
1
.723
5
2
6
.239
5
7
4
.381
6
2
3
.809
The
pr
op
os
ed
CSA
m
et
ho
d
is
te
ste
d
on
la
rg
e
-
scal
e
te
st
syst
em
s
with
30
,
60
,
and
10
0
gen
erati
on
un
it
s.
t
max
is
sel
ect
ed
to
be
10
00
0.
The
disp
at
ch
resu
lt
s
fo
r
the
60
-
un
it
te
st
syst
em
with
MFO
are
giv
en
in
Table
5.
The
op
ti
m
al
value
of
fu
el
cost
ob
ta
ined
by
the
pr
op
os
ed
CSA
fo
r
the
60
-
un
it
te
st
syst
em
is
37
42
.9
$/h
r.
The
resu
lt
s
of
la
rg
e
-
scal
e
te
st
syst
em
s
with
30
,
60
,
and
10
0
gen
era
ti
on
un
it
s
are
com
par
ed
to
QP
-
ALH
N
[3
]
,
CGA
[2
2],
and
IG
A
-
AMUM
[2
2].
The
com
par
ison
resu
lt
s
and
total
execu
ti
on
ti
m
e
are
giv
en
in
Table
6.
Fr
om
this
ta
ble,
we
ob
serv
ed
that
the
pr
op
os
ed
CSA
giv
es
the
app
ro
xim
at
e
resu
lt
s
as
QP
-
ALH
N
[3
]
fo
r
al
l
syst
em
s,
an
d
it
giv
es
bette
r
resu
lt
s
than
CGA
[2
2]
and
IG
A
-
AMUM
[2
2].
The
con
ver
gen
ce
char
act
erist
ic
of
fu
el
cost
ob
j
ect
ive
fu
nctions
fo
r
the
la
rg
e
-
sca
le
syst
em
s
is
sh
ow
n
in
Figu
re
2
.
Also
,
the
capab
il
it
y
of
the
pr
op
os
ed
CSA
m
et
ho
d
is
te
ste
d
fo
r
so
lving
ver
y
la
rg
e
-
scal
e
te
st
syst
em
s
with
50
0,
15
00
,
20
00
,
and
25
00
un
it
s.
The
best
fu
el
cost
resu
lt
s
and
the
total
execu
ti
on
are
giv
en
in
Table
7.
Fr
om
this
ta
ble,
it
is
ob
serv
ed
that
with
increasing
the
siz
e
of
the
te
st
syst
em
,
the
pr
op
os
ed
CSA
giv
es
bette
r
values
of
f
uel
cost co
m
par
ed
to
QP
-
ALH
N
[3
]
.
Table
5.
Disp
at
ch results
for 6
0
-
unit
test
syst
e
m
w
it
h
MF
O
Un
it
Fu
el
P
i
(
MW)
Un
it
Fu
el
P
i
(
MW)
Un
it
Fu
el
P
i
(
MW)
Un
it
Fu
el
P
i
(
MW)
1
2
2
1
8
.2489
16
3
2
3
9
.6314
31
2
2
1
8
.2485
46
3
2
3
9
.6321
2
1
2
1
1
.6634
17
1
2
8
8
.5852
32
1
21
1
.6610
47
1
2
8
8
.5849
3
1
2
8
0
.7220
18
3
2
3
9
.6314
33
1
2
8
0
.7226
48
3
2
3
9
.6320
4
3
2
3
9
.6316
19
3
4
2
8
.5343
34
3
2
3
9
.6315
49
3
4
2
8
.5193
5
1
2
7
8
.4958
20
1
2
7
4
.8658
35
1
2
7
8
.4992
50
1
2
7
4
.8677
6
3
2
3
9
.6308
21
2
2
1
8
.2516
36
3
2
3
9
.6307
51
2
2
1
8
.2
5
2
1
7
1
2
8
8
.5850
22
1
2
1
1
.6643
37
1
2
8
8
.5843
52
1
2
1
1
.6635
8
3
2
3
9
.6324
23
1
2
8
0
.7218
38
3
2
3
9
.6316
53
1
2
8
0
.7224
9
3
4
2
8
.5167
24
3
2
3
9
.6315
39
3
4
2
8
.5328
54
3
2
3
9
.6318
10
1
2
7
4
.8691
25
1
2
7
8
.4937
40
1
2
7
4
.8681
55
1
2
7
8
.4961
11
2
218.
2
4
7
6
26
3
2
3
9
.6314
41
2
2
1
8
.2504
56
3
2
3
9
.6316
12
1
2
1
1
.6634
27
1
2
8
8
.5852
42
1
2
1
1
.6629
57
1
2
8
8
.5820
13
1
2
8
0
.7230
28
3
2
3
9
.6316
43
1
2
8
0
.7233
58
3
2
3
9
.6313
14
3
2
3
9
.6311
29
3
4
2
8
.5073
44
3
2
3
9
.6306
59
3
4
2
8
.5210
15
1
2
7
8
.4981
30
1
2
7
4
.8683
45
1
2
7
8
.4962
60
1
2
7
4
.8650
Total Fu
el cos
t
3
7
4
2
.9
Table
6.
C
om
par
iso
n of t
he
m
in. fuel c
os
t
for
la
rg
e
-
scal
e tes
t sy
stem
Metho
d
No
.
o
f
un
its
Min. total f
u
el cos
t
Execu
tio
n
ti
m
e
(se
c)
Prop
o
sed
CSA
30
60
100
1
8
7
1
.4
0
0
3
7
4
2
.9
0
0
6
2
3
8
.1
0
0
2
1
.04
4
0
.55
7
2
.57
QP
-
A
LHN
[
3
]
30
60
100
1
8
7
1
.4
2
6
3
7
4
2
.8
5
5
6
2
3
8
.0
9
2
0
.13
0
.24
0
.43
CGA [
2
2]
30
60
100
1
8
7
3
.6
9
1
3
7
4
8
.7
6
1
6
2
5
1
.4
6
9
2
6
3
.64
5
1
7
.88
8
7
3
.70
IGA
-
AMU
M
[
2
2]
30
60
100
1
8
7
2
.0
4
7
3
7
4
4
.7
2
2
6
2
4
2
.7
8
7
8
0
.47
1
5
7
.19
2
7
5
.67
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
4
,
A
ugus
t
2020
:
3431
-
3440
3438
Figure
2
.
Co
nverg
e
nce c
ha
rac
te
risti
c
s o
f
t
he pr
opos
e
d
C
SA
for
la
r
ge
-
scal
e
te
st sy
stem
s
Table
7.
Re
s
ults o
f
the
b
e
st f
ue
l cost
f
or
ver
y
larg
e
-
scal
e tes
t sy
stem
Metho
d
No
.
o
f
un
its
Total f
u
el cos
t
Execu
tio
n
ti
m
e
(se
c)
Prop
o
sed
CSA
500
1500
2000
2500
3
1
1
9
1
.00
0
9
3
5
7
2
.00
0
1
2
4
7
6
0
.00
1
5
5
9
5
0
.00
3
2
0
.40
9
5
0
.55
1
3
0
2
.2
0
1
5
0
0
.4
6
QP
-
A
LHN
[
3
]
500
1500
2000
2500
3
1
1
9
0
.46
0
9
3
5
7
1
.37
0
1
2
4
7
6
1
.83
1
5
5
9
5
2
.29
9
.67
2
1
7
2
.828
3
7
5
.781
6
7
6
.563
5.3. C
as
e
3: P
ie
cew
ise
qu
adratic
fu
el
c
os
t
w
ith valve
-
p
oint
l
oa
di
ng ef
fe
cts
In
this
cas
e,
th
e
VP
L
is
c
on
si
der
e
d
al
on
g
wi
th
MFO.
The
capab
il
it
y
of
t
he
pro
pose
d
CSA
to
s
olve
this
prob
le
m
is
te
ste
d
on
the
10
-
un
it
te
st
sy
stem
with
27
00
M
W
loa
d
de
m
and
.
t
max
is
s
el
ect
ed
to
be
1000
0.
The
total
e
xec
ution
tim
e
is
15
.
70
sec.
T
he
disp
at
c
h
res
ult
s
an
d
the
c
om
par
is
on
res
ults
are
giv
e
n
i
n
T
able
8.
Fr
om
this
ta
bl
e,
it
is
noti
ced
that
only
pro
po
s
ed
C
SA,
S
SA
[4
]
,
DSD
[4
]
,
C
GA
-
MU
[24],
CS
A
[25],
a
nd
BSA
[
26
]
sat
isfy
powe
r
balance
c
on
st
rain
t,
an
d
the
othe
r
m
e
tho
ds
H
CR
O
-
D
E
[
4],
CB
PSO
-
R
VM
[23],
QP
S
O
[27],
a
nd
N
PS
O
-
LRS
[27
]
vio
la
te
it
.
Als
o,
it
is
no
ti
ced
that
the
value
of
m
in
fu
el
co
st
is
inc
rease
d
from
62
3.8
092
$/hr
i
n
case
1
for
2700
M
W
to
62
3.834
2
$/hr
in
this
case
due
t
o
th
e
VPL.
T
he
sta
ti
sti
cal
resu
lt
s
of
the
pro
posed
CS
A
m
et
ho
d
are
c
om
par
ed
to
I
GA
-
MU
[
24
]
,
C
G
A
-
M
U
[24],
a
nd
CS
A
[25],
a
s
gi
ve
n
in
Table
9.
It
cl
ear
that
the
pro
po
se
d
CS
A
gi
ves
bette
r
m
in
and
m
ean
val
ues
of
f
ue
l
cost
com
par
ed
t
o
the o
t
her m
et
h
od
s
.
Table
8.
Res
ults o
f 10
-
unit
tes
t sy
stem
w
it
h
pi
ecewise
qua
drat
ic
co
st f
unct
ion
and VPL c
om
par
in
g wit
h othe
r
al
gorithm
s
Uni
t
F
u
e
l
P
rop
o
sed
C
S
A
S
S
A
[4
]
HCRO
-
DE
[4
]
DSD
[4]
C
B
P
S
O
-
R
VM
[23
]
CGA
-
MU
[24]
QPS
O
[27
]
NPS
O
-
L
R
S
[2
7
]
C
S
A
[2
5]
B
S
A
[26
]
1
2
2
1
8
.8548
2
1
9
.16264
2
1
3
.4589
2
1
8
.59400
2
1
9
.2073
2
2
2
.0108
2
2
4
.7063
2
2
3
.3352
2
1
9
.1817
2
1
8
.57
2
1
2
1
2
.4086
2
1
1
.65928
2
0
9
.7300
2
1
1
.71174
2
1
0
.2203
2
1
1
.6352
2
1
2
.3882
2
1
2
.1957
2
1
1
.6596
2
1
1
.21
3
1
2
8
1
.5418
2
8
0
.68427
3
3
2
.0143
2
8
0
.65706
2
7
8
.5456
2
8
3
.9455
2
8
3
.4405
2
7
6
.2167
2
8
0
.6571
2
7
9
.56
4
3
2
3
9
.0244
2
3
9
.95493
2
3
7
.7581
2
3
9
.63943
2
7
6
.4120
2
3
7
.8052
2
8
9
.6530
2
8
6
.0163
2
3
9
.9551
2
3
9
.50
5
1
2
8
0
.1966
2
7
6
.3875
0
2
6
9
.1476
2
7
9
.93452
2
7
4
.6470
2
8
0
.4480
2
8
3
.8190
2
8
6
.0163
2
7
6
.4164
2
7
9
.97
6
3
2
3
9
.6657
2
3
9
.79532
2
3
8
.9677
2
3
9
.63943
2
4
0
.5797
2
3
6
.0330
2
4
1
.0024
2
3
9
.7974
2
3
9
.7953
2
4
1
.11
7
1
2
8
7
.4733
2
9
0
.07417
2
9
2
.3267
2
8
7
.72749
2
8
5
.5388
2
9
2
.0499
2
8
7
.8571
2
9
1
.1767
2
9
0
.0985
2
8
9
.79
8
3
2
3
9
.9521
2
3
9
.82117
2
3
7
.7557
2
3
9
.63943
2
4
0
.6323
2
4
1
.9708
2
4
0
.6245
2
4
1
.4398
2
3
9
.8207
2
4
0
.57
9
3
4
2
6
.0197
4
2
6
.37501
4
1
3
.6294
4
2
6
.58829
4
2
9
.4008
4
2
4
.2011
4
0
7
.9870
4
2
9
.2637
4
2
6
.3626
4
2
6
.88
10
1
2
7
4
.8632
2
7
6
.08571
2
6
6
.3841
2
7
5
.86861
2
7
6
.1815
2
6
9
.90
05
2
7
8
.2120
2
7
8
.9541
2
7
6
.0531
2
7
2
.79
T
P
G
2
7
0
0
.0
0
2
7
0
0
.0
0
2
7
1
1
.1
7
2
5
2
7
0
0
.0
0
2
7
3
1
.3
6
5
2
7
0
0
.0
0
2
7
4
9
.6
9
2
7
6
4
.4
1
1
9
2
7
0
0
.0
0
2
7
0
0
.0
0
F
u
e
l
c
o
st
6
2
3
.8342
6
2
3
.6433
6
2
8
.9605
6
2
3
.8265
6
2
4
.3911
6
2
4
.7193
6
2
4
.1505
6
2
3
.9258
6
2
3
.8361
6
2
3
.9016
0
100
200
300
400
500
600
700
800
900
1000
1000
2000
3000
4000
5000
6000
7000
I
t
e
ra
t
i
o
n
s
F
u
e
l
C
o
s
t
(
$
/
h
r)
3
0
u
n
i
t
s
6
0
u
n
i
t
s
1
0
0
u
n
i
t
s
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
So
lv
in
g
pract
ic
al econ
om
ic
l
oad dis
patc
h pr
ob
le
m usin
g
cr
ow se
ar
c
h alg
ori
thm
...
(
Shai
maa R.
Spea
)
3439
Table
9.
Stat
ist
ic
al
r
esu
lt
s fo
r
10
-
un
it
test
sys
tem
w
it
h
MFO
and
VP
L
Metho
d
Min f
u
el cos
t
Mean f
u
el cos
t
Max f
u
el cos
t
Std
Prop
o
sed
CSA
6
2
3
.8342
6
2
3
.8566
6
8
0
.0601
0
.62
9
0
CGA
-
M
U [
2
4
]
6
2
4
.7193
6
2
7
.6087
6
3
3
.8652
-
IGA
-
MU
[
2
4
]
6
2
4
.5178
6
2
5
.8692
6
3
0
.8705
-
CSA [
2
5]
6
2
3
.8361
6
2
3
.96
26
6
2
4
.8304
0
.01
1
6
6.
CONCL
US
I
O
N
In
this
pa
per,
t
he
CS
A
m
et
hod
has
bee
n
su
c
cessf
ully
i
m
plem
ented
to
s
olve
the
non
-
c
onve
x
pr
act
ic
al
ELD
pro
blem
with
valve
-
point
loa
ding
e
ff
ect
s
a
nd
m
ul
ti
-
fu
el
op
ti
ons
.
The
10
-
un
it
t
est
syst
e
m
has
bee
n
consi
der
e
d.
In
add
it
io
n
to
la
r
ge
-
scal
e
te
st
syst
e
m
s
with
30
-
unit
,
60
-
unit
,
and
10
0
-
unit
,
and
ver
y
la
r
ge
-
scal
e
te
st
syst
e
m
s
with
500,
1500,
2000,
an
d
2500
unit
s.
Thr
ee
dif
fere
nt
cases
are
eff
ic
ie
ntly
st
ud
ie
d.
The
sim
ulati
o
n
res
ults
co
nf
i
rm
the
rob
us
tness
a
nd
e
ff
ec
ti
ven
ess
of
th
e
propose
d
C
SA
m
et
ho
d
to
so
l
ve
the
pract
ic
al
ELD
prob
le
m
with
differe
nt
f
orm
ulati
on
s.
G
ood
c
onve
r
gen
c
e
char
act
e
risti
cs
of
t
he
CS
A
m
et
ho
d
is
detect
ed.
Th
e
si
m
ulati
on
resu
lt
s
are
com
par
ed
to
the
re
porte
d
al
gorith
m
s.
The
com
p
ariso
n
of
re
su
l
ts
and
the
sta
ti
sti
c
al
a
naly
sis
confir
m
the
eff
ect
ive
ness,
high
-
qual
it
y
so
luti
on
s,
a
nd
s
uperi
or
it
y
of
th
e
pro
pose
d
CSA
for
s
olv
i
ng the
pr
act
ic
al
EL
D pro
blem
.
REFERE
NCE
S
[1]
R.
Haba
chi
,
e
t
al.
,
"Ec
onom
ic
and
emiss
ion
di
spatc
h
using
cu
ckoo
sea
r
ch
a
lg
orit
hm
,
"
Inte
rna
ti
onal
Journ
al
of
El
e
ct
rica
l
and
C
omputer
Engi
n
e
ering
(
IJE
CE)
,
v
ol.
9
,
no
.
5
,
pp
.
3
384
-
3390,
Oc
t
.
2
019.
[2]
Y.
V.
K.
Redd
y
and
M.
D.
R
ed
d
y
,
"G
re
y
W
olf
Optimiza
ti
o
n
for
Solving
Ec
ono
m
ic
Dispatc
h
wi
th
Multi
pl
e
Fuel
s
and
Valve
Poi
n
t
Loa
ding
,
"
Int
e
rnational
Journ
al
of
Informa
ti
o
n
Engi
ne
ering
and
El
e
ct
ronic
Busine
ss
,
vol.
1
,
pp.
50
-
57
,
2019
.
[3]
Vo
N.
Dieu
and
P.
Schegn
er,
"A
ugm
ent
ed
La
g
ran
ge
Hopfi
el
d
net
work
in
it
i
al
i
z
ed
b
y
qu
adr
atic
progra
m
m
ing
fo
r
ec
onom
ic
dispatch
with
piece
wis
e
quadr
atic
cost
func
ti
ons
and
pr
ohibi
te
d
zone
s,"
A
ppli
ed
So
ft
Co
mputing
,
vol.
13
,
pp.
292
–
301
,
20
13.
[4]
J.
J.Q.
Yu
and
V.
O.
K.
Li
,
"A
Socia
l
Spider
Algorit
hm
for
S
olvi
ng
the
Non
-
conve
x
Ec
onom
ic
Loa
d
Dispat
c
h
Problem,"
Neuro
computi
ng
,
Jul.
2015.
[5]
L
d.
S.
Coe
lho
a
nd
V.
C.
Mariani
,
"Com
bini
ng
of
cha
ot
ic
diff
er
ent
i
al
evo
lut
ion
and
quadr
a
ti
c
progra
m
m
ing
for
ec
onom
ic
dispatch
opti
m
iz
a
ti
on
with
val
ve
-
poi
nt
eff
ect,
"
IEEE
Tr
an
sacti
ons
on
Powe
r
Syste
ms
,
vol.
21,
no.
2,
pp.
989
-
996
,
2
006.
[6]
J.
K.
Delson
and
S.
M.
Shahide
h
pour,
"Lin
ea
r
pr
ogra
m
m
ing
appl
ic
a
ti
ons
to
powe
r
s
y
stem
e
cono
m
ic
s,
pla
nning
a
n
d
oper
ations,”
IEEE
Tr
ansa
ct
ions
on
Powe
r S
ystem
s
,
vol. 7, no. 3, pp. 1155
–
1163,
Aug
.
1992.
[7]
N.
Am
ja
d
y
and
H.
Nasiri
-
Rad,
"
Soluti
on
of
nonconve
x
and
non
sm
ooth
ec
onom
ic
dispat
ch
b
y
a
new
ada
pti
v
e
re
al
code
d
g
enetic al
gorit
hm
,
"
E
xpe
rt
Syste
ms
wi
th Ap
pli
cations
,
vo
l. 3
7,
pp
.
5239
–
524
5,
2010
.
[8]
B.
R.
Adarsh
,
et
al.
"E
conomic di
spatc
h
using
ch
a
oti
c
bat algorit
h
m
,
"
Ene
rgy
,
vol
.
96,
pp.
666
-
675
,
2016
.
[9]
B.
Ben
toua
t
i
,
et
al.
,
"E
le
ph
ant
Herdi
ng
Optimi
za
t
ion
for
Solvi
ng
Non
-
conve
x
Optimal
Pow
er
Flow
Problem,
"
Journal
of
Elec
t
rical
and
Elec
tronic
s E
ng
ine
erin
g
,
vol
.
10
,
no
.
1
,
pp.
31
-
40
,
2017
.
[10]
M.
Modiri
-
Dels
had
and
N.
Ab
d
Rahi
m
,
"M
ulti
-
objecti
v
e
ba
ck
tra
ck
ing
sea
r
ch
al
gorit
hm
for
e
conomic
emiss
i
on
dispat
ch
problem
,
"
Appl
i
ed
Sof
t
Computing
,
vol
.
40,
pp.
479
–
494
,
2016
.
[11]
A.
A.
El
sak
aa
n
,
e
t
al
.
,
"E
conomic
Pow
er
Dispatch
with
Emiss
ion
Constrai
nt
and
Valve
Point
Loading
Eff
e
ct
Us
in
g
Moth
Flame
Opt
imiza
ti
on
Algori
thm,"
Adv
an
ce
d
Engi
ne
ering
For
um
,
vol
.
28
,
pp
.
139
-
149,
Jun.
20
18.
[12]
A.
As
kar
za
d
eh,
"A
novel
m
etah
eur
isti
c
m
et
hod
for
solving
constra
in
ed
eng
ineer
ing
opti
m
i
za
t
ion
proble
m
s:
Cro
w
sea
rch
al
gor
it
hm
,
"
Computers an
d
Struct
ures
,
vol
.
169
,
pp
.
1
–
12
,
2016.
[13]
A.
A.
Abou
El
El
a
,
et
al.
,
"
Appl
ic
ati
on
of
the
C
row
Searc
h
Al
gorithm
for
Ec
onomic
Env
ironmental
Dispatch
,
"
2017
Ninet
ee
nth
Inte
rna
ti
onal
Middle
ea
st
power
Sy
stems
Confer
enc
e
(MEPCO
N),
Menoufia
Univer
sity
,
Egy
pt,
Dec
.
2017.
[14]
B.
Z.
D
r
-
As
li
,
et
al.
,
"
Chapt
er
14:
Crow
Sear
ch
Algorit
hm
(CSA
)
,
"
Adv
ance
d
Optimizati
on
by
Nature
-
Inspired
Al
gorithms
,
Jul
.
2017
[15]
S
P.
S.
Bhull
ar
and
J.
K.
Dha
m
i,
"P
art
ic
le
Sw
arm
Optimiza
ti
on
Based
Ec
onom
ic
Loa
d
Dispatc
h
with
Valve
Point
Loa
ding,
"
Inte
rnational
Journal
of
Engi
nee
ring R
ese
arch
&
Technol
ogy
(
IJE
RT)
,
vol.
4
,
no.
5,
2015.
[16]
R.
M.
Riz
k
-
Al
l
ah
,
e
t
al
.
,
"
A
New
Sine
Cosine
Optimization
Algorit
hm
for
Solvin
g
Com
bine
d
Non
-
Conv
ex
Ec
onom
ic
and
Emiss
ion
Pow
er
Dispatc
h
Probl
ems
,
"
Inte
rnatio
nal
Journal
on
Ene
rgy
Con
ve
rs
ion
(
I.
R.E.
CON.
)
,
vol.
5
,
no
.
6
,
No
v
.
2017
.
[17]
R.
M.
Riz
k
-
Alla
h
,
et
al.
,
"
A
novel
Para
ll
el
hurric
ane
opti
m
iz
at
ion
al
gorit
hm
for
sec
ure
emiss
ion/
ec
on
om
ic
loa
d
dispat
ch
soluti
on
,
"
Appl
ie
d
Soft
Computing
,
vol.
63,
pp.
206
-
222,
2018.
[18]
C.
E.
Li
n
and
G.
L.
Vivia
ni,
"H
ie
rar
chi
ca
l
ec
onom
ic
dispat
ch
for
pie
ce
wise
quadr
at
ic
cost
func
ti
ons,"
IEE
E
Tr
ansacti
ons on
Powe
r A
pparatus and
Syste
ms
PA
S
-
103
,
pp.
1170
–
1175,
1984.
[19]
J.
-
B.
Park
,
et
al.
,
"A
Parti
cl
e
Sw
arm
Optimiza
ti
on
for
Ec
onom
ic
Dispatc
h
with
Nons
m
ooth
Cost
Functi
ons,"
IEE
E
Tr
ansacti
ons on
Powe
r Sy
stems
,
vol.
20,
no.
1,
2005.
[20]
K.Y.
Le
e
,
et
al.
,
"A
dapt
ive
Hopfiel
d
neur
al
net
works
for
ec
onom
ic
loa
d
dispat
ch,
"
IEE
E
Tr
ansacti
ons
on
Powe
r
Syste
ms
,
vol.
13,
pp.
519
–
526,
1998.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
10
, No
.
4
,
A
ugus
t
2020
:
3431
-
3440
3440
[21]
S.
Baskar
,
et
al.
,
"H
y
brid
gene
ti
c
al
gorit
hm
soluti
on
to
ec
onom
ic
dispat
ch
proble
m
,
"
Inte
rnational
Journal
Computers E
le
ct
rical
Engi
nee
ring
,
vol.
39,
no.
3,
pp.
407
-
419,
May
2003.
[22]
C.
L.
Chia
ng
an
d
C.
T.
Su,
"A
dapt
iv
e
-
improved
gene
tic
al
gori
th
m
for
the
ec
onom
ic
dispat
ch
of
unit
s
with
m
ult
ipl
e
fue
l
op
ti
ons,"
C
y
berne
tics and
S
y
stems
,
vol
.
36
,
p
p.
687
–
704
,
200
5.
[23]
C.
L.
Chia
ng
an
d
C.
T.
Su,
"A
dapt
iv
e
-
improved
gene
tic
al
gori
th
m
for
the
ec
onom
ic
dispat
ch
of
unit
s
with
m
ult
ipl
e
fue
l
op
ti
ons,"
C
y
berne
tics and
S
y
stems
,
vol
.
36
,
p
p.
687
–
704
,
200
5.
[24]
C.
-
L.
Chi
ang,
"I
m
prove
d
Gene
tic
Algorit
hm
for
Pow
er
Ec
onom
ic
Dispatc
h
of
Units
with
Valve
-
Point
Eff
ec
ts
an
d
Multi
ple Fuel
s,"
IEE
E
Tr
ansacti
o
ns on
Powe
r
Sys
te
ms
,
vol
.
2
0
,
no
.
4
,
pp
.
1690
-
16
99,
Nov
.
2005.
[25]
F.
Moham
m
adi
,
H.
Abdi,
"A
m
o
difi
ed
cro
w
sea
r
ch
al
gor
it
hm
(MCS
A)
for
solving
loa
d
dispatch
proble
m
,
"
Appl
i
ed
Soft
Computing
,
vol.
71
,
pp
.
51
-
6
5,
2018
.
[26]
M.
Modiri
-
Dels
had
and
S.
Hr.
A.
Kaboli
,
"Ba
c
ktra
ck
ing
sea
rch
al
gorit
h
m
for
solving
ec
onom
i
c
dispat
ch
probl
e
m
s
with
va
lve
-
poin
t
eff
e
ct
s
and
m
ultiple fue
l
opt
ions,
"
Ene
rgy
,
vol
.
1
16,
pp
.
637
-
649
,
2016.
[27]
AI
Selva
kum
ar
and
K.
Th
anushkodi,
"A
new
pa
rti
cle
sw
arm
opt
imiza
ti
on
soluti
o
n
to
nonconv
ex
ec
onom
ic
disp
atch
proble
m
s,"
IEEE
Tr
ansacti
on
on
Powe
r Sy
st
em
,
v
ol.
22
,
no
.
1
,
pp
.
42
-
51,
2007
.
Evaluation Warning : The document was created with Spire.PDF for Python.