TELKOM
NIKA
, Vol.12, No
.4, Dece
mbe
r
2014, pp. 77
3~7
7
8
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v12i4.439
773
Re
cei
v
ed Au
gust 12, 20
14
; Revi
sed O
c
t
ober 2
9
, 201
4; Acce
pted
No
vem
ber 1
2
,
2014
Cost Forecastin
g Model of Transmissio
n
Project based
on PSO-BP Method
Yan Lu
1,
Don
g
xiao Niu
2
, Bingjie Li
3
, W
e
idong Liu
4
1,2,
3
School of E
c
onom
ics and
Mana
geme
n
t, North Ch
ina El
ectric Po
w
e
r U
n
iversit
y
, No.2,
Beino
ng R
oad
,
Huil
on
ggu
an,
Cha
ngp
in
g Dis
trict, 102206, B
e
iji
ng, Ch
ina, P
h
./F
ax: 617
73
0
7
9
4
State Grid Z
hejia
ng El
ectric Po
w
e
r C
o
mpa
n
y
Ec
onom
ic R
e
searc
h
Institute, No.1, Nanfu
Roa
d
,
Shan
gch
eng D
i
strict, 31000
8, Z
heji
ang, Ch
in
a
e-mail: h
d
lu
ya
n
@
16
3.com
1
, niud
x@
126.com
2
, crazyjen
n
y
l
e
e@1
63.com
3
A
b
st
r
a
ct
In order to sol
v
e bei
ng sens
i
t
ive to the initi
a
l w
e
ights, slo
w
convergenc
e, bein
g
easy to fall into
local
min
i
mu
m and
oth
e
r pro
b
le
ms
of the
BP ne
ural
net
w
o
rk, this pap
er intro
duc
es the Partic
le Sw
a
r
m
Optimi
z
a
t
i
o
n
a
l
gorit
hm into t
h
e Artificia
l
Ne
ur
al N
e
tw
ork trainin
g
, an
d const
r
uct a BP n
eur
al n
e
tw
ork mo
d
e
l
opti
m
i
z
e
d
by th
e particl
e sw
arm opti
m
i
z
a
t
io
n.
T
h
is
metho
d
can spe
ed u
p
the conver
genc
e
and i
m
prov
e the
pred
iction
acc
u
racy. T
h
rou
gh
the a
nal
ys
is
of the
main
facto
r
s on t
he c
o
st
of trans
missi
on
lin
e
proj
ect, di
g
out the
p
a
th
and
le
ad
facto
r
s, topogr
ap
hy
an
d
met
eoro
l
ogic
a
l fact
ors, the t
o
w
e
r an
d the
tow
e
r b
a
s
e
mater
i
als
a
nd
other factors.
Use th
e PSO-
BP mod
e
l for
the c
o
st foreca
sting
of trans
mission
li
ne
pro
j
ect
base
d
o
n
hist
orical
pro
j
ect
data. T
he r
e
s
u
lt show
s
that
the metho
d
can
pr
edict
the cost effectiv
ely.
Co
mp
ared
w
i
th
the tra
d
iti
ona
l
BP ne
ura
l
n
e
tw
ork, the
metho
d
ca
n
pred
ict w
i
th h
i
gh
er
accu
racy, an
d c
a
n
be
gen
eral
i
z
e
d
an
d app
lie
d in co
st forecasting o
f
actual proj
ects.
Ke
y
w
ords
: PSO, BP neural n
e
tw
ork, 110kV transmissio
n
li
ne, cost foreca
sting
1. Introduc
tion
To meet the
so
cial de
ma
nd for el
ect
r
i
c
ity, the power g
r
id a
r
ou
n
d
ha
s bee
n
c
o
ns
tr
uc
te
d
mu
c
h
fa
s
t
er
.T
h
e
in
fr
as
tr
uc
tu
r
e
in
ves
t
me
n
t
o
f
th
e Sta
t
e
G
r
id C
o
r
p
or
a
t
io
n is
maintaine
d
a
n
ann
ual g
r
owth
rate of
over 1
0
%.A rea
s
o
nable
determi
natio
nto the
co
st of
con
s
tru
c
tion
proje
c
ts is i
m
porta
ntto improve
th
e
returns
of the po
wer gri
d
investm
ent
. At
pre
s
ent, it’s
mainly throu
g
h
the bud
get quota shall to
estimate a
c
curately the p
r
oject cost [1], but
this m
e
thod
ha
s b
een
incre
a
sin
g
l
y
unabl
e t
o
me
et the
re
quiremen
t
s of
econ
omic
development.In the c
ontex
t of not c
o
mpletely c
o
llec
t
ing t
he amount of informat
ion, it’s
hard
to
predi
ct the
p
r
oje
c
t cost q
u
ickly and
efficiently.
The
r
efore, the in
trodu
ct
ion of
advan
ced cost
forecastin
g method
s an
d the
improvementof cost pre
d
icti
o
n
accuracy
have impo
rtant
signifi
can
c
e.
Many schol
ars and
expert
s
laun
ch
ed a
studyin
the fi
eld of po
wer
engin
eeri
ng cost, but
mainly co
nce
n
trated in fa
ctors affe
cting
cost,
cost
co
ntrol an
d ma
nagem
ent an
d other a
s
p
e
cts,
relat
i
v
e
ly
f
e
w
e
r st
u
d
ie
s on
t
he co
st
f
o
re
ca
st
ing
model. Literature [
2
] us
es
the fuz
z
y
math theory,
and estim
a
te the co
st of th
e proje
c
t to be buil
tthrou
g
h
calculatingth
e
clo
s
e deg
re
e betwee
n
the
compl
e
ted
project
s
an
d th
e proje
c
ts to
be built;
litera
t
ure [3] ad
opt
s a li
nea
r reg
r
essio
n
mo
de
l to
predi
ct the
cost; literatu
r
e
[4] holds
a regre
s
si
on
an
alysis
on the
key imp
a
ctiv
e facto
r
s
on
the
co
st, usin
g m
u
ltiple linea
r regre
s
sion
an
d factor
adju
s
tment to e
s
tablish a com
p
reh
e
n
s
ive cost
forecastin
g m
odel fo
r the t
r
ansmi
ssion
p
r
oje
c
t; liter
atu
r
e [5] u
s
e
s
th
e GM
(1,1) m
odel to
esta
bl
ish
two pri
n
ci
ple
s
calcul
ation
model
s, wh
ich was
use
d
to com
p
ile
the estimate
s of the po
wer
engin
eeri
ng p
r
oje
c
ts.In the
appli
c
ation of
artificial
ne
ural netwo
rk, literatu
r
e [6] usesthe
co
st da
ta
of histo
r
ical p
o
we
r e
ngine
e
r
ing
proj
ect
s
for
ANN traini
ng, and
ado
p
t
s the n
e
w A
NN
after train
i
ng
to the
co
st fo
recastin
g of
n
e
w
po
wer pro
j
ects;
lite
r
atu
r
e [7] propo
se
d an
app
ro
ach ba
se
d on
the
combi
n
ing m
e
thod of gray relational
an
alysis
a
nd th
e neu
ral net
work; literatu
r
e [8] propo
se
d a
co
st fore
ca
sting method
ba
sed o
n
BP neural net
wo
rk
of the transmi
ssi
on line p
r
oj
ects.
Ho
wever, the
BP neural
netwo
rk algo
rithm
exist
s
probl
em
s ofb
e
ing
sen
s
itive to the
initial wei
ghts, easy to fall
into local minimum
an
d sl
ow
conve
r
ge
nce [9], so we introd
uce t
h
e
PSOalgorith
m
based oni
d
eas of glo
bal
stocha
stic
op
timization, an
d adopt the P
S
O-BP algo
rithm
tothe tran
smi
ssi
on line
co
st foreca
sting.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 773
– 778
774
2.
Analy
s
is of
The In
fluenc
ing Fac
t
ors
of
T
h
e T
r
an
smission Pr
oject
Co
st Based
on
The
Fishbone Diagram
The tra
n
smission li
ne p
r
oj
ect cost i
s
rel
a
ted to
ma
ny factors [10].
The
co
st per
length i
s
clo
s
ely relat
ed to the
voltage level
s
, terrai
n
, we
ather, to
we
rs, wire
s,
stee
l, con
c
rete a
n
d
earth
wo
rk,
etc. F
r
om
the
perspe
c
tive
o
f
co
st fo
re
ca
sting, extractt
he m
a
in fa
ct
ors affe
cting
the
level of its costba
se
d on
a ne
w tra
n
smissi
on lin
e
of ce
rtain vol
t
age level th
e, sho
w
e
d
a
s
the
Figure 1:
(1) Path an
d wire fa
ctor
s
The p
a
th le
ngth di
rectly
affects th
e
amount
of wire an
d to
we
rs vol
u
me, there
b
y
affecting the
material
a
c
quisitio
n
cost
s an
d
con
s
truction
co
sts;
the choi
ce
of wire mo
d
e
l is
determi
ned
b
y
the tra
n
spo
r
t capa
city of
the lin
e, an
d
the p
r
oje
c
t
costa
c
count
s
about
20% of
its
body
co
st, indire
ctlyaffects the i
n
fra
s
tructu
re
proje
c
tsco
stsof th
e tower
pa
rt [9]. The p
a
p
e
r
sele
ctsth
e
si
ngle line len
g
t
h, wire volu
me and wi
re
price as ind
e
x
to reflect the path and wires
f
a
ct
or
s.
Figure 1. Analysis of the i
n
fluen
cing fa
ctors
of the transmi
ssion p
r
oje
c
t of certa
i
n voltage
level based o
n
the fishbo
n
e
diagram
(2) T
opog
ra
p
h
ical a
nd met
eorol
ogi
cal fa
ctors
Different te
rrain di
re
ctly affects th
e e
a
se
of tra
n
smission
line
engi
neeri
ng
co
nst
r
uctio
n
,
tower-b
a
sed
form and
hu
man tran
sp
ort distance.Th
e terrai
n
is di
vided into hill
s, mountai
no
us,
mountain
s
, m
a
jesti
c
mou
n
tains a
nd
slou
gh.This
pap
e
r
com
b
ine
s
th
e increa
se
d costs
coeffi
cie
n
t
of variou
s types of terrain
and terrain a
c
counte
d
given in the qu
ota to cal
c
ulate
a
com
p
re
hen
siv
e
value, and
u
s
e
s
the inte
g
r
ated te
rrain
coeffici
ent
to
sho
w
the
effect of top
ography on
co
st
. In
addition, met
eorol
ogi
cal fa
ctor is al
so o
ne of t
he important facto
r
s affecting th
e co
st of power
transmissio
n
lines,
wind
a
nd ici
ng
hav
e speci
a
l re
quire
ment
s to the
sele
cti
on of
wire
s
and
towers
.
(3) T
o
we
r an
d tower b
a
s
e
material fa
cto
r
s
The dete
r
min
a
tion of towe
r sel
e
ctio
n, towe
r an
d other mate
rial
s shoul
d con
s
ider the
path of th
e li
ne, voltage
levels, the
nu
mber of
lo
op
s, terrain,
we
ather
and
ot
her fa
cto
r
s,
and
these
acco
un
t for a la
rge
prop
ortio
n
of
the co
st
of th
e line e
ngin
e
ering
body. If the sel
e
ctio
n
of
transmissio
n line co
rrid
o
r become
s
m
o
re difficult
a
n
d
the tortuous
path
c
o
efficient inc
r
eases
, i
t
will cause increase
s of strain, corner towers
and material consum
ptions, as a result, the body
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Co
st Fore
ca
sting Model of Tran
sm
issi
on
Project ba
se
d on PSO-BP
Method (Yan
Lu)
775
investmentin
crea
se
s, and
therefo
r
e, the numb
e
r
of
strain a
nd
corne
r
towe
r is one of t
he
importa
nt factors affe
cting
co
st.
(4) Oth
e
r fa
ctors
Earthwork an
d con
c
rete a
m
ount influe
nce o
n
the constructio
n
costs of
tran
smissi
on
lines
a lot. Other m
a
terial
s
use
d
in po
we
r tran
sm
i
ssi
o
n
line
s
,
su
ch as
the steel consumption a
nd
price alsoaffe
ct great o
n
its cost an
d are
t
he criti
c
al se
ctionof con
s
truction
work costs.
3. The Cos
t
Foreca
sting
Model of Tra
n
smission Project
Base
d
on The PSO-BP Me
thod
3.1 The Ba
sic Models
(1) BP neu
ral
networ
k
BP neural
n
e
twork i
s
a
multilayer f
eed-fo
rward
netwo
rk t
r
ai
ned by e
r
ro
r ba
ck-
propagation algorithm.It’s widel
y used and has a strong ge
nerali
z
ation ability and fault
tolerance.
BP neural
net
work can l
e
arn and
st
ore input - output rel
a
tionship m
apping.It’s
learni
ng
rule i
s
the ste
epe
st
desce
nt method, an
d
throug
h the
back-pro
pag
ation, con
s
ta
ntly
adju
s
ting th
e
wei
ghts an
d
thre
shol
ds o
f
the net
wo
rk, so th
at the
sq
uared
error
rea
c
h
e
s the
minimum [11]
-[12].
(2) Pa
rticle
swarm optimization algo
rith
m
The PSO first is to gen
erate a fea
s
ible sol
u
tion,
and then t
he obje
c
tive
function
determi
ne a
fitness val
u
e
[13].Each p
a
rticle
will m
o
ve in the
solution
spa
c
e
,
and spee
d
will
determi
ne its direction and distan
ce [14]. Typically,
the particl
es
will follow the current opti
m
al
particl
e, an
d throu
gh
se
arching e
a
ch g
e
neratio
n to
fin
d
the o
p
timal
solutio
n
. In e
a
ch
gen
eratio
n,
the parti
cle
s
will tra
ck it
s optimal soluti
on f
ound
so
far and th
e two extre
m
e
s
of the optim
al
s
o
lution found s
o
far [15].
3.2 The PSO-BP Hy
brid
Algorithm a
nd Its Implementa
tion
From
a poi
nt of mathem
a
t
ical view, th
e BP algo
rithm natu
r
eta
k
es the
erro
r
sum
of
squ
a
re
s
as t
he o
b
je
ctive functio
n
, an
d
finds th
e mi
ni
mum
with th
e g
r
adie
n
t m
e
thod.The
r
ef
ore,
the squa
red
error fu
nctio
n
is
po
sitively definit
e, oth
e
rwi
s
e
the
r
e
must exi
s
t a
local minim
u
m
points; the P
S
O algorith
m
essentially b
e
long
s to
a ra
ndom o
p
timization process, and there i
s
no
local
co
nverg
ence p
r
oble
m
s.The fiel
d
whe
r
e th
e PSO
is m
o
st
widely u
s
e
d
i
s
optimi
z
atio
n [16].
Therefore,
co
mbinethe PS
O and BP ne
ural n
e
two
r
k,
and the
step
sof this algo
rithm (PSO-BP
)
is
s
h
ow
n
in
F
i
gu
r
e
2
.
4. Case Stud
y
4.1 Parameters Setting
In ord
e
r to ve
rify the validity of PSO-BP algor
ithm us
ed
in
co
s
t
for
e
c
a
s
t
in
g
o
f
th
e p
o
w
er
transmissio
n line proje
c
ts, the pape
r sel
e
cts
43 sets
of data of 11
0kV lon
g
line
(> 1
k
m
)
proje
c
ts
in Zhejian
g
p
r
ovince.We
select 34
sets of data as
th
e training
sa
mples, a
nd the re
st of the 10
sets a
s
te
st
sample
s, an
d
use
the BP
al
gorithm
an
d t
he PSO
-BP a
l
gorithm
top
r
e
d
ict, an
d
ca
rri
es
on the co
ntra
st analysi
s
.
In this pap
er,
a BP neural
netwo
rk
with
3 laye
rs is u
s
ed: the numb
e
r of neu
ro
nsin input
layer is set to
14, and they arethe1
4 factorsscre
ene
d in these
c
on
d se
ction; the hidden laye
r is
determi
ned
b
y
trial a
nd
error m
e
thod,
a
nd eve
n
t
ually
identified
29
neuron
s; the
numbe
r
of o
u
t
pu
t
layer is 1, an
d it’s the co
st per length of
eac
h p
r
oje
c
t
.The transfe
r function u
s
e
d
in the hidd
en
layer i
s
lo
gsi
g
, and
the
pu
relin i
n
the
o
u
tput layer.
T
he training
fu
nction
is train
gdm, the
trai
ning
times of
network is 10
000,
and th
e e
r
ror
of trainin
g
g
o
als i
s
1e-5. T
he n
u
mbe
r
of parti
cle
s
i
s
2
0
,
the evolution
times is 20, the maximum
allowabl
e
nu
mber of iterat
ions is 2
0
, the accele
ratio
n
con
s
tant
s c1 i
s
2.8 and
c2 i
s
1.3, and the
maximum sp
eed limit of Vmax is1.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 773
– 778
776
Figure 2.Steps of the PSO-BPalgorithm
4.2 Resul
t
s
Before th
e
sa
mple d
a
ta i
s
i
nput to th
e n
eural
net
wo
rk, norm
a
lize t
he d
a
ta a
c
co
rding
to
the formula (1).
m
in
m
ax
m
in
'
XX
X
X
X
(1)
Figure 3
an
d Figu
re
4
are th
e the
test sample
fitting maps
for tra
n
smi
ssion line
engin
eeri
ng
with BP al
go
rithm an
d PS
O-BP al
gorith
m
. Figu
re
5 i
s
the
comp
arison
chart
of
the
predi
cted val
ue and the
a
c
tual value of
transmi
ssion
line based o
n
BP algorith
m
and PSO-BP
algorithm. T
h
is paper
select
s
the mean absol
u
te
percenta
ge error
(M
APE)to judge the
forecastin
g effects, and the
formula is:
1
1
n
ii
n
i
OD
M
APE
ND
(2)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Co
st Fore
ca
sting Model of Tran
sm
issi
on
Project ba
se
d on PSO-BP
Method (Yan
Lu)
777
Figure. 3 The
test sampl
e
fitting map
based on BP algorith
m
Figure. 4 The
test sampl
e
map ba
sed o
n
PSO-BP algorithm
Figure. 5 The
compa
r
i
s
on
map of the fore
castin
g and
actual value
based on BP and PSO-
BP algorithm
The detaile
d result
s are
sh
own in the Ta
ble 1:
Table.1 Th
e fore
ca
sting an
d actual valu
e based on B
P
and PSO-B
P algorithm
Project No.
Actual Value
Forecasting Value
(BP)
Forecasting Value
(PSO
-BP
)
Relative Error
(BP)
Relative
Error
(PSO
-BP
)
35 83.69
84.44
83.79
0.90%
0.12%
36 63.13
59.80
56.00
5.27%
11.29%
37 44.37
48.91
44.32
10.25%
0.10%
38 68.90
66.84
68.92
2.99%
0.03%
39 86.06
87.26
86.13
1.39%
0.08%
40 43.11
43.39
48.60
0.65%
12.74%
41 80.01
64.74
79.99
19.08%
0.02%
42 107.19
113.52
107.15
5.91%
0.03%
43 67.31
73.19
73.86
8.74%
9.74%
44 47.11
51.81
47.10
9.98%
0.02%
Usi
ng the equation (2)
we can obtain t
hatthe MAPE of
the BP algorithm i
s
6.52%, and
3.42%ba
sed
on PSO
-BP algorith
m
.This i
ndi
cate
s that the
PSO-BP mo
del ha
s
obv
ious
advantag
es o
n
improvin
gth
e
forecastin
g accuracy.
5. Conclusio
n
(1)
This pa
pe
r d
i
gs an
d anal
yses the m
a
in fact
orsaffe
cting the cost of the transmissio
n line
proje
c
tba
s
e
d
on the fish
bo
ne diag
ram f
r
om fo
u
r
a
s
p
e
cts
of the p
a
th and le
ad,
topography
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 12, No. 4, Dece
mb
er 201
4: 773
– 778
778
and meteo
r
ol
ogical factors, the tower an
d tower b
a
se material
s an
d
other facto
r
s.
(2)
This pa
pe
r combine
s
the
particle
swa
r
m optimization algo
rithm
and BP neural net
work,
prop
osi
ng th
e PSO-BPforeca
s
ting
mo
del.Throug
h
the a
c
tual m
easure
m
ent
of the p
o
we
r
transmissio
n l
i
ne p
r
oje
c
ts i
n
Zheji
ang P
r
ovince,
we
ge
t the co
st of t
he a
c
tual
eng
ineeri
ng.Th
e
predi
cted
re
sults sh
ow th
at, the appli
c
ation of PS
O-BP model
in transmission lin
e co
st
forecastin
gha
s goo
d ap
pl
ication effe
cts of
fast co
nverge
nce speed a
nd hi
ghpredi
ction
ac
cur
a
cy
.
(3)
The p
r
e
d
ictio
n
mod
e
l PSO
-BP ca
n u
s
e
the gl
obal
se
a
r
ch
ability to
optimize
the i
n
itial weight
s
value, so th
a
t
can
solve t
he p
r
o
b
lem
s
of se
nsit
ive to the initial
state,
easy
to
fall into local
optimal value
and
slo
w
co
nverge
nce ra
te of BP neu
ral n
e
two
r
k.
The al
gorith
m
ha
s some
advantag
es,
and can be
widely use
d
in tran
smi
ssi
on li
ne proj
ect co
st fore
ca
sting
.
Referen
ces
[1] Qi
Liu.
T
heory
and Meth
ods o
f
Project Cost Mana
ge
me
nt
. Beiji
ng: Ch
in
a Electric Po
w
e
r Press. 2004.
[2] Yi
Fan.
F
u
zz
y
Mathe
m
atic
al-
base
d
Proj
ect Cost Estimate
on T
r
ans
miss
io
n Li
ne
. North
Chin
a El
ectri
c
Po
w
e
r U
n
ivers
i
t
y
. 201
2.
[3]
Heh
ong Gu
o. Rese
arch o
n
c
o
st forecastin
g
of t
he transmis
s
ion pr
oj
ect ba
sed o
n
regr
ess
i
on A
nal
ys
is.
Shanx
i Electric
Pow
e
r.
2008; 36(3): 75-
77.
[4]
Jiao
W
ang,
Le
qun
Di
ng. C
o
m
p
reh
ensiv
e
pre
d
ictio
n
mo
dels
for transmissi
o
n
e
ngi
ne
erin
g
proj
ect cost
base
d
on its ke
y infl
ue
ncin
g fa
ctors.
East China Electric Power.
2009; 36(
11): 111-
11
3.
[5]
Z
hen
xi
n W
ang.
T
he GM (1,1) Mode
l in C
o
st F
o
recastin
g of Electric Po
w
e
r Constructi
on P
r
oject.
Inne
r
Mong
oli
a
Elect
r
ic Pow
e
r T
e
chnol
ogy.
19
96; (5): 46-50.
[6]
Qiang Yi
ng, Yi
ng T
an, Pengj
un Chi. T
he Cost F
o
re
castin
g of the Electric Po
w
e
r Proje
c
t based on
ANN.
Journ
a
l o
f
Jiangxi Voc
a
ti
ona
l an
d T
e
ch
nical C
o
l
l
eg
e o
f
Electricity.
2012; 24(4): 2
1
-2
3.
[7]
Yongm
ing Ya
n
g
, Yan W
ang,
Xiu
j
u
n
F
an, Cha
o
Liu. Cos
t
Estimate of
Po
w
e
r Li
ne b
a
s
ed on Gre
y
Relati
on
al An
al
ysis a
nd N
eura
l
Net
w
orks.
Jo
urna
l of Cho
n
g
q
in
g Univ
ersity
.
2013; 11: 15-
20.
[8]
Yunp
en
g L
i
ng,
Pen
g
fei Y
an,
Cha
ngzh
an
Ha
n, Ch
e
n
g
u
a
ng
Yang. T
he C
o
s
t
F
o
recastin
g
Mode
l of th
e
T
r
ansmission L
i
ne Pro
j
ect bas
ed on BP.
Elec
tric Power.
2012; 45(10): 9
5
-9
9.
[9] Qiongs
hu
ai
Lv.
BP Neural N
e
tw
ork Optimi
z
a
t
i
on a
nd Rrse
ar
ch
. Z
hengzh
o
u
Universit
y
. 20
11.
[10]
Z
hen
yu Z
h
ao,
Qianl
ei lv, W
e
i
y
a
ng Y
ou, Yu
n
Gao. An Eva
l
u
a
tion In
de
x M
o
del
of Co
nstruc
tion C
o
st for
35kV T
r
ansmission L
i
nes i
n
R
u
ral Po
w
e
r Net
w
o
r
k.
Pow
e
r System T
e
ch
nol
ogy
. 200
8; 32(
14): 96-1
00.
[11]
W
u
Junke, Z
h
ou L
u
o
w
e
i
, D
u
Xi
on
g, Sun P
eng
jun. J
uncti
on T
e
m
peratur
e Pred
iction
of
IGBT
Po
w
e
r
Modu
le Base
d
on BP Neura
l
Net
w
o
r
k.
Jour
nal of el
ectrica
l
eng
ine
e
rin
g
& technol
ogy.
201
4; 9(3):
970-
977.
[12]
Budi
Ra
hman
i
,
Supri
y
a
d
i. E
a
rl
y Mo
del
of
T
r
affic Sign Remi
nd
er B
a
sed
on
Ne
u
r
al N
e
t
w
ork.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
ng.
2012; 1
0
(4): 7
13-7
2
2
[13]
A. Ismail, DS. Jeng, L
L
. Z
han
g. An optim
ise
d
pr
o
duct-un
it neur
al n
e
t
w
ork
w
i
t
h
a n
o
ve
l P
S
O-BP h
y
bri
d
traini
ng al
gorit
hm: Applic
atio
ns to load-
def
ormatio
n
ana
l
y
sis of a
x
i
a
ll
y load
ed pi
les.
Engin
eeri
n
g
app
licati
ons of
artificial i
n
tel
lig
ence
. 26(
10): 2
305-
231
4.
[14]
JJ. Jamian, M
W
. Mustafa, H. Mokhlis. A N
e
w
P
a
rticle S
w
a
rm Optimization
T
e
chniqu
e in
Optimizin
g
Size of Distributed Generation.
Internatio
nal
Journa
l of Ele
c
tr
ical an
d Co
mp
uter Eng
i
n
e
e
rin
g
.
201
2;
2(1): 137-
14
5.
[15]
Lep
ing T
an, Jun W
en. Optimi
zation of the B
P
Neural N
e
t
w
ork
w
i
t
h
Particl
e
S
w
a
rm Algor
ithm.
Journa
l
of Hube
i Univ
e
r
sity for Nation
alities (N
atura
l
Scienc
e Editio
n).
2012; 3
0
(3)
:
254-27
0.
[16]
Na Li, Yu
xi
a Li
. Electricit
y
Pri
c
e
F
o
recast ba
sed on PSO-B
P Neura
l
Net
w
ork.
Engin
eer
in
g Journ
a
l o
f
W
uhan U
n
iver
sity.
2008; 41(
4): 102-1
05.
Evaluation Warning : The document was created with Spire.PDF for Python.