TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 14, No. 3, June 20
15, pp. 525 ~ 5
3
3
DOI: 10.115
9
1
/telkomni
ka.
v
14i3.785
0
525
Re
cei
v
ed Fe
brua
ry 8, 201
5; Revi
se
d April 18, 201
5; Acce
pted Ma
y 8, 2015
A Weighted Evidence Combination Method Based on
Improved Conflict Measure Factor
Xiaochen Xing*
1
, Yuan
w
e
n Cai
2
, Zhe
ng
y
u
Zhao
3
, Long
Ch
eng
4
, Yan Li
5
1,3
Department of Graduate M
ana
geme
n
t, Equipm
ent Acad
e
m
y
,
Beij
in
g, Ch
ina 1
0
1
416
,
(+
86)1
500
13
62
5
7
2
2,4,
5
Department
of Space Equ
i
pment, Equi
pm
ent Academ
y,
Beiji
ng, Ch
in
a 101
41
6
,
(+
86)150
01
362
57
2
*Corres
p
o
ndi
n
g
author, em
ail
:
xin
g
x
ia
och
e
n
_ht@1
63.com
1
, cai
y
uan
w
e
n
@
26
3.net
2
A
b
st
r
a
ct
D-S evi
denc
e
theory is
usu
a
l
l
y use
d
for th
e fusio
n
of
multi-so
urce i
n
fo
rmati
on. But t
he fusi
o
n
result is
alw
a
ys aga
inst w
i
th ge
ner
al
k
n
o
w
ledge for
th
e he
avy co
nfl
i
ct of evid
enc
e. Rese
arch
on
combi
natio
n of
conflict ev
id
en
ce at h
o
m
e
an
d abr
oa
d is
s
u
mmar
i
z
e
d
an
d
ana
ly
z
e
d
in
det
ail. On th
e b
a
s
e
of
this, the concl
u
sion that
mo
difi
ed ev
i
d
e
n
ce co
mb
in
ation
method of co
nflic
t
evid
ence
is mo
re useful c
an b
e
daw
n. Effective evid
enc
e co
nflict meas
ur
e
is the first step of conflict ev
i
denc
e co
mb
in
ation. T
he ex
is
ting
conflict
me
asur
e metho
d
s are
summari
z
e
d
a
nd the
main
pr
obl
e
m
of those
meth
ods is
an
aly
z
e
d
in
detai
l
.
Based
o
n
pr
e
v
ious
rese
arch
of co
nflict
evi
denc
e co
mbi
n
ation, a modifi
ed meas
ur
e f
a
ctor of
evi
d
e
n
ce
conflict w
h
ic
h
i
s
cal
l
ed
Mco
n
f is
put forw
ard
.
Mconf is
mai
n
ly bui
lt
u
p
w
i
th mo
difie
d
dist
ance
of evid
en
c
e
na
me
d md
BPA
and tra
d
iti
ona
l
evid
enc
e co
n
f
lict factor na
me
d
k. T
he
e
x
ampl
es in
thi
s
pap
er sh
ow
that
Mconf can
me
asure the ev
id
ence co
nflict correctly, both for gen
eral
evid
ence a
nd co
nfli
ct evidenc
e.
Ke
y
w
ords
:
conflict
evid
en
ce co
mbin
atio
n, confl
i
ct me
as
ure, modifi
e
d
evi
d
e
n
ce di
stance,
trad
iti
ona
l
conflict factor, impr
ove
d
confli
ct factor
Copy
right
©
2015 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
Due
to the
differen
c
e
s
of
knowl
edge
a
c
quisi
tio
n
ap
proache
s an
d t
he me
asurin
g
error of
the sy
stem
a
nd the
sen
s
o
r
, there
are
redun
dan
cie
s
and
contrad
i
ctorie
s
in m
u
ltiple sou
r
ce
s.
Demp
ste
r-Sh
a
fer theo
ry is an effect
ive method to fuse un
ce
rtainti
e
s
of co
nflict informatio
n. It is
signifi
cant
effect u
s
in
g
D-S
eviden
ce
the
o
ry to
solve
the u
n
certain
probl
em
s b
e
cause of th
e la
ck
of kno
w
le
dg
e. With the i
n
temsive
stu
d
y of D-
S e
v
idence theo
ry, it’s found
that there
are
perv
e
rs
e c
o
n
c
lu
sion
s
whe
n
f
u
sin
g
mult
i
-
so
ur
ce
ba
se
d on the
D-S
eviden
ce the
o
ry [1]. Aiming at
this p
r
oble
m
, resea
r
chers
have cond
uct
ed in d
epth rese
arch both at
home and abro
ad,
an
d put
forwa
r
d
so
m
e
sol
u
tion
s.
The p
epe
r
wi
ll con
dut a
d
e
tailed a
naly
s
is. Ba
se
d o
n
the
critici
s
m o
f
pred
ecesso
rs’ research
re
sults,
an imp
r
o
v
ed confli
ct measure fact
or M
c
onf ba
sed on
corre
c
ti
on
distan
ce
md
BPA
and conflict factor
k
is p
u
t forwa
r
d. The co
nflict be
tween evide
n
c
e
s
is mea
s
u
r
ed
based on
md
BPA
and the
weight value
of
ea
ch
eviden
ce
i
s
cal
c
ulat
ed. The
evid
ence i
s
mo
dified
and the final
deci
s
io
n ca
n be obtain
ed b
a
se
d on the
modified evid
ence.
2. Induction
of Con
f
lict E
v
idence Fusion
Aiming at
the
fusio
n
of con
f
lict eviden
ce,
it can
be
ro
u
ghly divide
d i
n
to two
g
r
ou
ps. T
he
first gro
up is
modified com
b
ination rule
met
hod, anot
her g
r
ou
p is to revise the e
v
idences.
2.1. The Modified Combination Rule Method
The metho
d
of modifying and revi
sing
rue
s
co
nsi
d
e
r
s that the ap
pearan
ce of cou
n
ter-
intuitive resul
t
is due to the use of disp
osa
b
le
app
ro
ach in ha
ndli
ng eviden
ce
confli
ct base
d
on
Demp
ste
r
co
mbination
rul
e
. There a
r
e
some
typical
example
s
su
ch a
s
Ya
ger
rule, Smet
s rule,
DP rule, Sun
Quan rule, PCR5 rule p
r
o
posed by Smaran
da
che a
n
d
De
sert.
Evidence
co
nflicts
were classified into
global X
wha
t
is com
p
lete
ly unkno
wn t
h
rou
g
h
Yager
rule [2
]. It can solv
e the proble
m
of fusi
on o
f
two
highly conflicting eviden
ce
s.
The low
sup
portin
g
e
v
idence still
sup
port
s
lo
w after fusi
on
throu
gh
su
ch treatme
nt. And the
conf
lict
eviden
ce i
s
compl
e
tely a
band
oned
by
su
ch treat
m
ent. The
r
efore, even if th
ere
are
multi
p
le
eviden
ce
s strongly suppo
rting the
focal
element of t
he previou
s
confli
ct evide
n
ce. Th
e fin
a
l
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 14, No. 3, June 20
15 : 525 – 53
3
526
fusion
re
sult
i
s
e
n
tirely
neg
ative and
irra
tional.
Smets rul
e
would
cl
assify se
ction
co
nflict to
th
e
empty s
e
t, cons
is
tent wit
h
the Yager rule’s
pr
oblem [3]. DP rule ass
i
gns
the BPA of c
onflic
t
eviden
ce to
u
n
ion
set of th
e conflict fo
cal elem
ent
[4]
.
It is
s
u
itable fo
r stro
ng co
nflict
eviden
ce,
while
it al
wa
ys seem
s
co
nse
r
vative a
nd the
conv
erge
ne
is sl
o
w
. Sun
Qu
a
n
rule
co
nsi
d
ers
confli
ct evidence i
s
still available whose degr
ee of
the availabl
e is
depended on credibili
ty
ε
defined
by Su
n Qua
n
rule [
5
]. When
de
a
ling with
high
confli
ct evide
n
ce, P
C
R5 ru
le ha
s a
ce
rta
i
n
advantag
e. While
deali
n
g with g
ene
ral non
-conflic
t evidence, the fusi
ng eff
e
ct is
bad
er
than
traditional De
mpste
r
rule’
s
conve
r
ge
nce
and PCR5 rul
e
is not asso
ciative [6-7].
2.2. The of Rev
i
sing Ev
id
ence Me
thod
Amend evide
n
ce
meth
od consi
ders De
mpste
r
co
m
b
i
nation rule
i
s
right.
Th
e ap
pearan
ce
of paradox i
s
due to
the e
rro
r of
evide
n
ce. Ev
ide
n
ce should
be
modified
befo
r
e
combi
natio
n,
then the
con
c
lusio
n
i
s
mo
re re
asona
nle
in the
aspe
cts
of physi
cs, mathemati
c
s and
l
ogi
c.
Th
ere
are typical e
x
amples
su
ch as Mu
rphy
aver
age evi
den
ce metho
d
, Deng yon
g
expectatio
n
eviden
ce met
hod, Yeqin
g
weig
hte evid
ence meth
od
, Liu Zhu
nga
relative
wei
ghting evid
en
ce
method, Yin xuezh
ong
we
ighted evide
n
ce meth
od,
Liuwei
ru co
nflict combi
n
ation re
cog
n
i
t
ion
method, Li
u
Zhung
a inte
g
r
ated
wei
ght
ed evide
n
ce
method
an
d Li b
o
inte
grated
weigh
t
ed
eviden
ce met
hod.
Murp
hy mad
e
the m
a
ss
of co
rrespon
ding to
th
e f
o
cal
elem
ent
of all the
e
v
idence
sha
r
ing
[8]. T
hen
co
mpo
s
ite n
-
1
eviden
ce
s b
a
sed
on
De
mp
ster
ru
le. The
meth
od d
eem
s
ea
ch
evidence equal wei
ght. While each source of
information has different reli
abilities, or t
he
sen
s
o
r
s m
a
y have failure i
n
pra
c
tical p
r
oblem
s, so e
a
ch evid
en
ce
shoul
d have
different wei
g
ht.
Den
g
yong computed Jo
u
s
selme
di
sta
n
ce bet
ween
two eviden
ces in
ord
e
r t
o
obtain
distan
ce mat
r
ix DM [9].
Evidence
si
milarity
matri
x
SM was d
e
fined 1-DM.
Then obtain
all
eviden
ce wei
ghts ba
sed o
n
simil
a
rity
m
a
trix.
T
he evi
den
ce
s a
r
e
given differe
nt
weig
hts a
nd t
hen
sum all of the
m
. Combin
e the modified e
v
idences ‘n
-1’
times ba
sed
on Dem
p
ste
r
rule.
Ye
Qing cal
c
ulate
conflict factor betwe
e
n
tw
o evid
en
ce
s to ge
nerate co
nflict
matrix K
[10]. K was normali
ze
d an
d take
s entro
py to gener
at
e evidence weight coeffi
cients. Modify the
eviden
ce ba
sed on the wei
ght factor. Co
mposit
e the
revised evid
en
ce ba
se
d on
Sunqua
n rul
e
.
Liu Zhu
nga
calcul
ated
cre
d
ibility as a
weight value
b
a
se
d on
De
n
g
Yong’
s pa
p
e
r. Ma
ke
the maximu
m wei
ght val
ue
corre
s
po
n
d
ing evid
en
ce a
s
sta
nda
rd eviden
ce,
and the
sta
n
dard
eviden
ce is n
o
t to be pro
c
essed [7]. Th
e remai
n
ing
eviden
ce
s are modified in
accordan
ce
with
weig
hts
com
parin
g
with standard evid
ence. The
su
rplu
s ma
ss v
a
lue of
amen
dable
eviden
ce
wa
s a
s
sign
ed
to full
set of
each evid
en
ce. Co
mbine
t
he mo
dified
e
v
idences ba
sed o
n
Demp
ste
r
rule.
Yin Xuezhon
g cal
c
ul
ate e
a
ch
wei
ght value b
a
sed o
n
Liu Zh
ung
a [11]. Do n
o
t sele
ct
standard evidence, while all ev
idences
will be revised.
Com
b
ine the
evidence based on
Demps
t
er rule.
Liu WeiRu propo
sed
se
cti
onal
confli
ct m
easure ba
sed on
gambi
ng p
r
omi
s
e
s
distan
ce
difBetP and conflict factor
k
[12]. It main
ly adopts the method of
thresh
old dete
r
mination. Do
not
operate a
n
ything
with difB
etP and
k
. Assign
a valu
e for the
co
nflict
thre
shol
d
ε
a
c
cordi
ng to th
e
pra
c
tical
appl
ication. If and
only if the value of difBetP and
k
are la
gre
r
than o
r
e
qual to
ε
value,
there
i
s
a serious
conflict betwe
en evidences.
In
the
re
st conditio
n
s,
combi
ne
eviden
ce b
a
se
d
on Dem
p
ste
r
rule. Liu Wei
R
u ju
st put forwa
r
d a
ki
nd of
compo
s
ite measur
e of evidence confli
ct
based on difB
etP and
k
. But it did not give a solutio
n
when the confli
ct is large.
The co
nflict of evidence
can’t be me
a
s
ur
ed prope
rl
y based on
eviden
ce dist
ance or
conf
li
ct
f
a
ct
or
k
only. Liu Z
hung
a put forwar
d the two’
s ge
ometri
c
mean
BP
A
kd
to rep
r
esent
confli
ct bet
we
en evid
en
ce
s [13]. Revise
the evide
n
ce
s afte
r
cal
c
ul
ating e
a
ch ev
iden
ce
wei
g
h
t
s
.
Comp
osite
th
e revi
sed
evi
den
ce
s
with
expectatio
n
s
eviden
ce m
e
thod a
nd
rel
a
tive wei
ghti
n
g
eviden
ce met
hod.
Li Bo pointe
d
out that using
BP
A
kd
to measu
r
e the de
gree of confl
i
ct betwe
en
eviden
ce
s, when the difference of
the e
v
idence is
small, the con
f
lict measure
value ba
sed
on
BP
A
kd
gro
w
too fa
st co
mpa
r
ed
with
BP
A
kd
. And it is ea
sy to
ca
use
erro
r cal
c
ulatio
n. But
whe
n
the value of n of con
f
lict measu
r
e
factor ta
ke
s too much, the confli
ct measure facto
r
will
be
not se
nsitive
to the ch
ang
e of eviden
ce
in a
certai
n
rang
e. So, L
i
Bo took
BP
A
kd
as confli
ct
measure fact
or [14].
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Weighted E
v
ide
n
ce Com
b
ination Meth
od Base
d on
Im
proved
Co
nflict… (Xiao
c
he
n Xing)
527
2.3. Compre
hensiv
e Ev
aluation of T
w
o Method
s
Demp
ste
r
rul
e
have
some
good
math
e
m
atical
prope
rties,
su
ch
a
s
co
mmutative
law an
d
asso
ciative l
a
w. Mo
dify the combi
nati
on rule
s
u
s
u
a
lly dest
r
oy
the mathem
a
t
ical p
r
op
erti
es
.
Whe
n
the
co
mbination
rul
e
can
not
sati
sfy the a
s
so
ci
ative law, m
u
l
t
iple evidence fusi
on
order is
boun
d to affect the fusion
result. If combining a
ll the
evidence together, the calcul
ation will
be
exploding. In
fact, aimin
g
at the p
r
obl
em of h
ealth
monito
ring
of com
p
lex
system, wh
en
the
sen
s
o
r
bre
a
ks do
wn or o
r
there is a tra
n
smi
ssi
on error, stro
ng co
nflict evidence will come o
u
t. It
is irratio
nal th
at to make the probl
em to combi
nat
ion rule [15]. Therefore it is more rea
s
on
able
to
combi
ne conf
lict eviden
ce
based on revising evid
en
ce
method.
3. Measure o
n
Ev
idence Con
f
lict
It’s requi
re
d that eviden
ce
s be m
u
tually
i
ndep
ende
nt whe
n
De
mp
ster
com
b
inat
ion rul
e
s
is a
pplied
to
the combin
ation of m
u
lti-source
eviden
ce,
so
eviden
ce
s a
r
e i
nde
pend
ent of
e
a
ch
other by def
ault. The
rul
e
that the
minority i
s
subordinate
to the
majo
ri
ty is ap
plied
to
Trou
ble
s
ho
oting evide
n
ce,
that is to
say,
if one
eviden
ce i
s
reje
cted
by othe
rs, it i
s
very li
kely t
o
be fault evid
ence, and it
s intensity
sh
ould b
e
wea
k
en
ed d
u
rin
g
eviden
ce
co
mbination. T
he
degree of op
positio
n amo
ng eviden
ce i
s
call
ed evid
ence confli
ct, so confli
ct betwee
n
evide
n
ce
sho
u
ld be me
asu
r
ed.
3.1. Curren
t
Metho
d
for the Meas
ure
of Con
f
lict
Curre
n
tly, there are 2 kin
d
s
of method t
o
m
easure the confli
ct of
evidence: one
of the
m
is
the conflict fac
t
or
k
p
r
o
posed
by De
mpste
r
a
nd
Shafer; the
o
t
her i
s
th
e
confiden
ce
lev
e
l,
whi
c
h i
s
ba
sed on
Jo
usselme di
stan
ce cal
c
ul
ati
on.
Origin
ally, Jousselm
e di
stance i
s
u
s
e
d
for
the calculatio
n of the
difference b
e
twe
e
n
the
de
ci
sio
n
by evid
ence combin
atio
n an
d the
re
a
lity.
Then it is use
d
to measu
r
e
the conflict betwee
n
the evidences. It is pointed that a single meth
od
of the two ca
n’t measu
r
e t
he co
nflict exactly in
pape
r [12] and [14]. Therefo
r
e, some re
se
arch
ers
prop
osed that
the deci
s
ion
be ba
sed o
n
k
and d in the
form of the two.
3.2. The Main Problem
Peng
pro
p
o
s
ed that,
Jo
usselme
di
stan
ce is in
su
fficie
n
t wh
en
it is a
pplied
to m
e
a
s
ure th
e
gene
ral evide
n
ce
confli
ct. The big
ger B
PA degre
e
of dispe
r
sion i
s
, the smalle
r the Jou
s
selm
e
distan
ce of t
w
o evide
n
ce. In fact, each gro
up of
e
v
idence is completely co
nflicted. As for 2
grou
ps of category evide
n
c
e, it is totally c
onflicte
d a
m
ong evide
n
c
e
s
but the Jousselm
e dist
ance
is not the
ma
x value 1. Th
e rea
s
o
n
is th
at whe
n
it is
norm
a
lized, the de
nomin
ator is con
s
t value
2, and
it
can’
t extend.
Hen
c
e,
a
m
en
dme
n
t eviden
ce
d
i
stan
ce
BP
A
md
is p
r
opo
sed.
On t
he b
a
si
s
of excellent
cha
r
a
c
ter
ret
ention, all th
e pro
b
lem
s
are
solved
a
nd the stypti
city is better.
Therefore, a
m
endm
ent e
v
idence dista
n
ce
BP
A
md
and
k
is combin
ed
to measure
evidence
confli
ct in this paper [16].
The definition
of amendme
n
t evidence d
i
stan
ce
BP
A
md
is as formula (1) a
nd (2
):
11
2
2
1
2
BP
A
11
2
2
,,
2
,
,,
mm
m
m
mm
md
mm
m
m
(
1)
12
nn
12
1
2
11
,(
)
(
)
ij
ij
ij
ij
A
B
AB
A
B
mm
m
m
(
2)
The produ
ct
of
k
and
BP
A
d
is u
s
ed in
both p
aper [1
3] and
[14] to measure the evid
e
n
ce
confli
ct, and t
he uni
que
dif
f
eren
ce i
s
th
e value
of n i
n
n
BP
A
()
kd
. It is poi
nte
d
in p
ape
r [1
3] that
when the BPA value c
hanges
,
k
a
nd
BP
A
d
have no dire
ct
correlatio
n. If the produ
ct rule for the
para
m
eter
BP
A
md
and
k is
ap
plie
d by the
tho
ught of
pap
e
r
[
13]
and
[1
4], the evide
n
ce
conflict
can’t be
m
e
a
s
ured exactly
.
The rea
s
on
is
that
o
n
ce a
paramete
r
i
s
0, 2 evid
en
ces
are
de
cid
e
d
not conflicte
d
no
matter th
e othe
r
pa
ra
meter i
s
.
Ho
wever,
fact
s
are
not
all
so
. As i
s
i
ndi
cat
ed in
example 1.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 14, No. 3, June 20
15 : 525 – 53
3
528
Ex 1 Suppose frame of discernme
n
t is
1
234
56
,,
,,
,
X
, t
he BPA of 2
evidenc
e
s
are:
ca
se1
:
11
2
3
4
(,
,
,
)
1
m
;
2
4
5
236
(
,
)0
.
8
(
,
)0
.
2
mm
ca
se 2
:
12
()
()
0
.
2
,
1
,
2
,
,
5
ii
mm
i
ca
se 3
:
11
2
1
2
3
(,
)
(
,
)
1
,
[
0
,
1
]
ma
m
a
a
;
21
mm
ca
se 4
:
11
1
2
(
)
0.99
(
)
0.01
mm
21
2
2
1
2
3
(
,
)0
.
0
1
(
,
,
)0
.
9
9
mm
31
3
2
()
0
.
9
(
)
0
.
1
mm
ca
se 5
:
1
(
)
1
/
6,
1
,
2,
,
6
i
mi
;
k
()
1
,
k
2
,
3
,
,
m
m
ca
se 6
:
11
1
2
()
0
.
9
(
)
0
.
1
mm
;
11
1
2
()
0
.
1
(
)
0
.
9
mm
The value
s
of traditional
confli
ct fact
or
k
、
ame
n
d
m
ent eviden
ce di
stan
ce
BP
A
md
an
d
BP
A
md
k
in the cases
of example 1 are indi
cate
d in Table 1.
Table 1. The
values of
k
a
nd
BP
A
md
and
BP
A
md
k
of example 1
Case
BP
A
md
k
BP
A
md
k
BP
A
()
/
2
km
d
Case 1
m
1
与
m
2
0.8729
0
0
0.4365
Case 2
m
1
与
m
2
0 0.8
0
0
Case 3
m
1
与
m
2
0 0
0
0
Case 4
m
1
与
m
2
0.8127
0
0
0.4064
m
1
与
m
3
0.0949
0.1080
0.0102
0.1015
m
2
与
m
3
0.7914
0
0
0.3957
Case 5
m
1
与
m
k
0.8452
0
0
0.4226
Case 6
m
1
与
m
2
0.8835
0.82
0.7245
0.8518
It can be co
n
c
lud
ed from t
he analy
s
is of
example 1:
In case 1, eviden
ce
m
1
and
m
2
are in conflict,
BP
A
md
k
misj
udge
s eviden
ce a
s
0.
In case 2 and
3, evidence
m
1
and
m
2
are identical, a
nd there i
s
n
o
confli
ct betwee
n
the
eviden
ce
s. No matter wh
a
t
the value ‘
a
’ is, the val
u
e of
BP
A
md
is al
wa
ys 0,
so
BP
A
md
k
ca
n
judge evide
n
c
e
s
in co
nflict
.
In ca
se 4,
eviden
ce
m
1
and
m
3
are si
gn
ificantly in
su
pport
of
1
, and
m
2
judge
s tha
t
the
possibl
e
con
c
lu
sion
may
be in
clu
ded
in
12
3
(,
,
)
with the
p
o
ssibility of 9
9
%. By analy
s
is, th
e
confli
ct between
m
1
an
d
m
3
is small
e
r than that be
tween
m
1
an
d
m
2
。
But in fac
t, the c
o
nflic
t
value bet
wee
n
m
1
an
d
m
2
is 0,
whi
c
h
i
s
smalle
r tha
n
that of
m
1
and
m
3.
So it is
cont
rary t
o
analysi
s
. Wh
at’s mo
re, the co
nflict deg
ree b
e
twe
en
m
1
and
m
2
are evidently di
fferent from t
hat
betwe
en
m
2
and
m
3.
It i
s
con
c
lu
ded
th
at there i
s
n
o
co
nf
lict b
e
tween
2 g
r
ou
ps of evide
n
ces by
confli
ct measurem
ent of
BP
A
md
k
and that the judgeme
n
t is wrong.
In ca
se
5, ev
iden
ce
m
1
in
dicate
s th
at
all the
eleme
n
ts in
identifi
c
ation
fram
e
can’t
be
assured. Eviden
ce
m
k
in
d
i
cate
s that id
entification i
s
tota
lly unkn
o
wn. Evidentl
y
, they are n
o
t
equal, ie, the
r
e is
confli
ct a
m
ong
evi
den
ce
s. Wh
en
co
nflict mea
s
u
r
ement of
BP
A
md
k
is a
pplied,
it is con
c
lud
e
d
that there is no confli
ct bet
wee
n
2 gro
ups of evide
n
c
e
s
, whi
c
h is
wro
ng.
In ca
se
6, o
b
v
iously, there
is
co
nflict b
e
twee
n evid
en
ce
m
1
and
m
2.
Value
BP
A
md
k
is bi
g
and the jud
g
e
m
ent is rig
h
t.
On the ba
sis
of the above analysi
s
, the fo
llowin
g
co
nclusio
n
s
can b
e
dra
w
n:
(1) A
s
ca
se 2
in example 1
,
when 2 evid
enc
es a
r
e ide
n
tical, there i
s
no conflict;
(2) As case
1 an
d
ca
se
3 in
exampl
e1, when
th
ere
is intersection
in
all
the focal
element
s of the 2 eviden
ces(i
n
terse
c
tio
n
is not
), the traditional
conflict facto
r
k
i
s
a
l
w
a
y
s
0
,
the ame
ndm
ent eviden
ce
distan
ce
BP
A
md
valu
e is
not al
wa
ys 0. In thi
s
case, th
ere
m
u
st n
o
t be
confli
ce between eviden
ce
s,
but
BP
A
md
k
value
is
always
0
by the
effect of
k
, whi
c
h
is not
c
o
rrec
t.
(3) A
s
ca
se 4
in exampl
e 1
,
in the situ
ati
on whe
r
e the
r
e a
r
e m
any (take 3
a
s
exa
m
ple
)
s
o
ur
ce
-e
vid
e
n
c
e
,
if a
ll the fo
c
a
l e
l
emen
ts
amo
n
g 2
different
evid
ences an
d th
e third evid
e
n
ce
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Weighted E
v
ide
n
ce Com
b
ination Meth
od Base
d on
Im
proved
Co
nflict… (Xiao
c
he
n Xing)
529
have interse
c
tion, conflict fator
k
is 0. But the 2 evidences a
r
e no
t equal to the third. So th
e
method with
BP
A
md
k
is
not c
o
rrec
t.
(4) As
ca
se 5
in example 1,
when there is eviden
ce a
s
“
()
1
m
”, other evid
ences have
no co
nflict wit
h
it, and method by
BP
A
md
k
is
not c
o
rrec
t.
3.3. Impro
v
e
m
ent of
Con
f
lict Mea
s
ure
Method
Combi
ne the
analysi
s
in 3.2 and the
achi
ev
ement
by prede
ce
ssors, it is pro
posed a
novel eviden
ce confli
ct measu
r
e facto
r
Mconf, it is expre
s
sed a
s
formula (3):
BP
A
BP
A
BP
A
00
Mcon
f
0
2
md
km
d
md
(3)
From form
ula
(3), it can be
concl
ude
d t
hat Mconf dep
end
s on tradi
tional confli
ct fator
k
and ame
ndm
ent eviden
ce
distan
ce
BP
A
md
. T
here a
r
e two
cases. First
,
when evide
n
ce
s are
idecti
cal, there is no
confli
ct and
confli
ct value
is 0; se
con
d
, whe
n
eviden
ce
s
are n
o
t identi
c
al,
take
th
em as equal
an
d su
m
the we
ig
hts rega
rdl
e
ss of
produ
ct
of
BP
A
md
and
k
in
pap
er [1
3] an
d
[14]. To guarantee the val
ue of Mconf b
e
twee
n [0,1], norm
a
lization
is nee
ded. A
s
k
a
nd
BP
A
md
is
betwe
en 0 an
d 1, the deno
minator i
s
2.
Once the nov
el confli
ct measu
r
e facto
r
Mconf
i
s
appli
ed to measure the confli
ct betwe
en
the eviden
ce,
there are ch
ara
c
ters a
s
follows:
(1)
12
Mc
onf
(
,
)
[
0
,
1
]
mm
;
(2)
12
Mconf
(
,
)
0
mm
,
if and on
ly if
12
m=
m
;
(3)
12
Mc
onf
(
,
)
1
mm
,
if and on
ly if
()
(
)
ij
AB
,
i
A
和
j
B
are fo
cal element
s o
f
m
1
and
m
2
;
The
ce
rtificati
on of
the
ch
aracte
rs ab
ove
is e
a
sy
and
it
is in
t
r
od
uced
simply i
n
thi
s
pape
r.
As
k
a
nd
BP
A
md
is between
0
and
1,
(1
) is
ce
rtificat
ed. If and
o
n
ly if
12
m=
m
,
BP
A
md
is
0
,
ch
ara
c
te
r (2) is
certificate. If and only if
k
and
BP
A
md
are
1,
12
Mconf
(
,
)
1
mm
, then it can be
con
c
lu
ded
()
(
)
ij
AB
, and the opp
osi
t
e is true. So cha
r
a
c
ter (3)
is ce
rtificate.
In actu
al a
p
p
licatio
n, wh
en the
novel
co
nflict m
e
asu
r
e
facto
r
Mconf i
s
a
pplied
to
measure the
conflict bet
wee
n
the eviden
ce, a thresh
old value
ε
should be
set with act
ual
situation. It can be id
entified that
there is big
confli
ct only whe
n
k
and
BP
A
md
are big.
Thre
sh
old
value
ε
is set as 0.7.
Measure the confli
ct in
the situation
s
like
example 1, it can be
con
c
l
uded:
As
ca
se
1 in
exampl
e 1,
eviden
ce
m
1
and
m
2
are small
,
12
Mconf
(
,
)
mm
=0.
4
3
6
5
,
so
judgem
ent is
right.
AS case 2 a
nd ca
se 3 in
example 1,
m
1
and
m
2
are identical, and there i
s
n
o
confli
ct,
12
Mconf
(
,
)
mm
=0
,
a
nd the j
udgem
ent is
corre
c
t.
As ca
se
4 in
example 1,
conflict amo
n
g
m
1
and
m
2
、
m
2
and
m
3
、
m
1
and
m
3
are small.
12
Mconf
(
,
)
mm
=0.
4
0
6
4
,
23
Mc
onf
(
,
)
mm
=0.3957
,
13
Mconf
(
,
)
mm
=
0
.1015
,
judgement is
right.
As case 5
in
example
1, whatever
k
i
s
,
confli
ct bet
we
en
m
1
a
nd
m
k
are
sm
all, i.e., any
evidenc
e
is
not in c
onflic
t
with
()
1
m
,
1
Mconf
(
,
)
k
mm
=0.
4
2
2
6
,
judgem
ent is co
rrect.
As ca
se 6 i
n
example 1,
there i
s
a bi
g confli
ct bet
wee
n
m
1
and
m
2,
12
Mconf
(
,
)
mm
=0.85
18, and
the judgeme
n
t is co
rre
ct.
In co
ncl
u
si
on
, the novel
co
nflict mea
s
u
r
e facto
r
M
c
o
n
f ca
n m
e
a
s
u
r
e th
e
conflict
amon
g
eviden
ce.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 14, No. 3, June 20
15 : 525 – 53
3
530
4. Weighted
Ev
idence Co
mbination b
ased on Mc
o
n
f
4.1. Algorith
m
Design
Assu
ming the
identification
framework
12
m
X,
,
,
, and the num
ber of eviden
ce is
n. Weighte
d
eviden
ce co
mbination al
g
o
rithm ba
se
d on Mco
n
f is d
e
sig
ned a
s
fo
llows:
a) See all th
e eviden
ce
s
as a g
r
ou
p, the co
nflict of
one evide
n
ce from the ot
her o
ne,
whi
c
h two are both fro
m
the group,
i
s
cal
c
ulate
d
ba
sed
on M
c
onf
. And the co
n
f
lict of every two
eviden
ce
s in the gro
up sho
u
ld be do
ne.
,
Mc
o
n
f
ij
is sh
own as fo
rmula (4).
,
,B
P
A
,
Mc
o
n
f
,
1
,
2
,
,
n
2
ij
ij
ij
km
d
ij
(4)
b) Evide
n
ce
confli
ct matri
x
is
con
s
tru
c
t
ed by
,
Mc
o
n
f
ij
, and t
he o
r
de
r
of
matrix is
nn
.
Con
f
is sho
w
n as form
ula (5).
1,
2
1
,
n
2,
1
2
,
n
n,
1
n
,
2
nn
0
c
onf
c
onf
c
onf
0
c
onf
=
0
c
onf
c
o
nf
0
Co
nf
(5)
c) Th
e co
nflict summation
of evidence
i
and all the ot
her evide
n
ce from the grou
p is
cal
c
ulate
d
ba
sed o
n
Con
f
(
i
), and
Co
nf
(
i
) is sho
w
n a
s
formul
a (6
).
n
,
1,
co
n
f
ij
jj
i
i
Co
nf
()
=
(
6
)
d) The
sup
p
o
r
t degree of e
v
idence
i
whi
c
h is
sup
port
ed by the other eviden
ce
s
from the
grou
p ca
n be
dra
w
n a
s
()
ui
whi
c
h is
sho
w
n a
s
formul
a (7
).
n
1
()
1
/
i
ui
i
i
Co
n
f
C
o
nf
()
(
)
(7)
e) The
weig
ht value of evidence
i
can b
e
drawn as
()
i
, which i
s
sh
own
as formul
a (8
).
n
1
()
(
)
/
i
iu
i
u
i
()
(8)
f) The
two
e
v
idences fro
m
the
gro
u
p
sh
ould
be
i
ndep
ende
nt
whe
n
u
s
e
Dempste
r
combi
nation
rule. T
he m
o
dified evid
en
ce i
n
p
ape
r [
13] an
d [14]
has a
sig
n
ificant co
rrelati
o
n
obviou
s
ly because of its ge
neratin
g meth
od. So Demp
ster
combi
nat
ion rule
can’t
be used in thi
s
situation [1
7]. In this p
ape
r, the combi
n
ation evide
n
ce
co
mb
m
can b
e
d
r
awn f
r
om th
e
weig
hted
eviden
ce co
mbination of
all t
he eviden
ce in the grou
p, and
co
mb
m
is sho
w
n a
s
formul
a (9).
n
co
m
b
1
()
i
i
mi
m
(9)
g) Assum
ed t
hat the num
b
e
r of focal el
ements i
n
evi
den
ce in
co
mb
m
is
k
, and the focal
element
s of
co
mb
m
can be call
ed
k
A
. The final
e
v
idence
fi
n
a
l
m
whi
c
h i
s
sho
w
n
as fo
rmula
(1
0)
can b
e
dra
w
n
from
co
mb
m
[18].
X
fi
na
l
c
om
b
,2
1
()
(
)
1
,
2
,
,
lk
k
lk
AA
k
mm
A
l
m
A
(10)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Weighted E
v
ide
n
ce Com
b
ination Meth
od Base
d on
Im
proved
Co
nflict… (Xiao
c
he
n Xing)
531
4.2. Verification of T
y
pic
a
l Case
In order to
make a go
o
d
comp
ari
s
o
n
, the example in pape
r [14] is sele
cted for
validation. Assuming th
e
identificatio
n frame
w
o
r
k
12
3
4
X,
,
,
, and the numbe
r of
eviden
ce
s is
6. The typical
ca
se
contai
n
s
two
sit
uatio
ns. In
situatio
n 1, the
sen
s
or i
s
no
rmal,
and
the conflict b
e
twee
n evide
n
ce
s i
s
small
.
In sit
uation
2, the
sen
s
or i
s
failu
re,
and th
e confl
i
ct
betwe
en evid
ences i
s
larg
e.
Situation 1
:
The sy
stem a
nd the sen
s
o
r
are b
o
th no
rmal, and BP
A of every eviden
ce is
s
h
ow
n
as
fo
llo
w
s
:
1
234
1
(,
,
)
0
.
9
,
(
X
)
0
.
1
mm
;
21
2
1
3
4
(,
)
0
.
1
,
(
,
)
0
.
9
mm
31
2
3
3
4
(,
)
0
.
2
,
(
,
)
0
.
8
mm
;
41
2
3
1
(
,
,
)
0.95
,
(
X
)
0.5
mm
54
52
3
4
()
0
.
6
,
(
,
,
)
0
.
4
mm
;
61
3
4
6
(
,
,
)
0
.
75
,
(
X
)
0.2
5
mm
The co
mbin
ation re
sult of evidences in
si
tuation 1 ba
sed on the pro
posed alg
o
rit
h
m in
this pap
er is
sho
w
n a
s
tabl
e 2. And the resu
lt can be
comp
ared wit
h
the algorith
m
s in pap
er
[14].
Table 2. The
combi
nation
result in situ
ation 1
Number
Combination
rule
s
BPA of evidence after
combination
Weight values
1 D-S
combination
rule
2
(
)
0.0028
m
4
(
)
0.9773
m
34
(
,
)
0
.0199
m
Null
2
Murph
y
average
evidence algorithm[8]
4
(
)
0.8296
m
34
(,
)
0
.
1
4
6
5
m
[0.1667, 0.16
67,
0.1667
0.1667, 0.16
67,
0.1667]
3
Liu Zhun
Ga com
p
rehensive
w
e
ight
ed eviden
ce
algorithm[13]
4
(
)
0
.
977
1
m
34
(
,
)
0
.0222
m
[0.62, 0.99, 1.
00
0.99, 1.00, 1.
00]
4
Li Bo comprehen
sive
w
e
ight
ed eviden
ce
algorithm[14]
4
(
)
0
.
977
1
m
34
(
,
)
0
.0200
m
[1.00, 1.00, 1.
00
1.00, 1.00, 1.
00]
5
Algorithm in this paper
1
(
)
0.
13
50
m
2
(
)
0.1657
m
3
(
)
0.2745
m
4
()
m
0.
424
8
[0.1687, 0.16
85,
0.1688
0.1613, 0.16
52,
0.1674]
Situation 2
: The system
i
s
no
rmal, an
d
eviden
ce
m
1
is differen
t
from it in
si
tuation 2
because
of the
sensor failure, the other
evidence have
no cha
nge. T
he
changed BPA
of
eviden
ce
m
1
is sh
own as fo
llows:
11
11
3
()
0
.
9
,
(
,
)
0
.
1
mm
The
com
b
ina
t
ion re
sult
of eviden
ce
s in
si
tuation
2 b
a
s
ed
on th
e p
r
opo
sed
algo
ri
thm in
this pap
er is
sho
w
n a
s
tabl
e 3. And the result al
so can
be comp
ared
with the algo
rithms in p
a
p
e
r
[14].
From typi
cal
ca
se of
situa
t
ion 1 a
nd
situat
ion 2,
we
can
co
ncl
ude
that no m
a
tter th
e
sen
s
o
r
i
s
in f
a
ilure
or not,
the propo
se
d
algo
rithm in
this pa
pe
r ca
n ma
ke a
de
cisi
on
co
rre
ct
ly.
Whe
n
the
r
e
i
s
a
sen
s
o
r
fai
l
ure, th
e fault
eviden
ce
ca
n be
cl
ea
rly i
dentified
ba
sed o
n
the
wei
g
h
t
values
of initial eviden
ce
s. And the n
egativ
e effects brought
b
y
t
he fault evidence can
be
eliminated to
maximize. Th
en the final correct de
ci
sio
n
is co
ncl
ude
d.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 14, No. 3, June 20
15 : 525 – 53
3
532
Table 3. The
combi
nation
result in situ
ation 1
Number
Combination
rule
s
BPA of evidence after
combination
Weight values
1 D-S
combination
rule
1
(
)
1
.
00
00
m
Null
2
Murph
y
average
evidence algorithm[8]
1
(
)
0.
09
94
m
4
(
)
0.7949
m
[0.1667, 0.16
67,
0.1667
0.1667, 0.16
67,
0.1667]
3
Liu Zhun
Ga com
p
rehensive
w
e
ight
ed eviden
ce
algorithm[13]
4
(
)
0.8868
m
34
(
,
)
0
.0250
m
14
(,
)
0
.
0
1
2
4
m
[0.49, 0.76, 0.
77
1.00, 0.73, 1.
00]
4
Li Bo comprehen
sive
w
e
ight
ed eviden
ce
algorithm[14]
4
(
)
0.9729
m
34
(,
)
0
.
0
2
0
1
m
[0.63, 0.99, 1.
00
1.00, 0.99, 1.
00]
5
Algorithm in this paper
1
(
)
0.
27
75
m
2
(
)
0
.
115
1
m
3
(
)
0.230
6
m
4
()
m
0.
376
8
[
0.1478
, 0.16
90,
0.1701
0.1720, 0.16
51,
0.1760]
5. Conclusio
n
In this pape
r, rese
arch o
n
combi
natio
n of
conflict
eviden
ce at home an
d a
b
roa
d
is
summ
ari
z
ed
and analy
z
ed
in detail. And we con
c
lu
de
that effective
evidenc
e
co
nflict measure is
the first step
of conflict evidence com
b
ination.
The
r
e are som
e
problem
s in
existing con
f
lict
measure met
hod
s, so a n
e
w c
onflict
measure fact
or Mconf is
prop
osed ba
sed o
n
previous
resea
r
ch. Th
e case
sho
w
that M
c
onf
can
mea
s
u
r
e
eviden
ce
co
nflict corre
c
tly. A weig
hte
d
eviden
ce co
mbination
alg
o
rithm ba
sed
on
M
c
onf
is
d
e
sig
ned, and
the
typical
ca
se sh
ow
that the
prop
osed al
gorithm
can
compl
e
te the eviden
ce
combi
nation
effectively,
both for ge
n
e
ral
eviden
ce an
d
conflict evide
n
ce.
Referen
ces
[1]
Guang
lei M
e
n
g
, Guang
ho
ng
Gong. W
e
ig
ht Coeffici
ents
Calcu
l
ati
on for
Evide
n
ce So
urces a
nd it'
s
Appl
icatio
n in Evide
n
ces
F
u
s
i
on.
J
ourn
a
l
of
Beiji
ng
Un
ivers
i
ty of Aer
o
n
auti
cs an
d Astro
n
a
u
tics
. 20
10
;
36(1
1
): 136
5-1
368.
[2]
Yager
RR. O
n
the
Dempst
er-Shafer F
r
am
e
w
o
r
k
and
N
e
w
Com
b
in
atio
n
Rules.
Infor
m
a
t
ion Sc
ienc
es
.
198
7; 41(2): 93
-137.
[3]
Smets P.
T
he
Combi
nati
on of
Evi
denc
e in the T
r
ansferab
l
e Beli
ef Model.
IEEE Transaction on Patter
n
and Mac
h
in
e Intelli
ge
nce
. 19
90; 12(5): 4
47-
458.
[4
]
D
u
bo
i
s
D
,
Pra
d
e
H
.
Re
p
r
ese
n
t
a
t
i
o
n
a
n
d
C
o
mbi
n
a
t
i
o
n
o
f
U
n
ce
rta
i
n
t
y
w
i
th
Be
li
e
f
F
u
n
c
ti
on
s and
Possib
ilit
y Mea
s
ures.
Co
mp
utation
a
l Intel
lig
e
n
ce
. 199
8; (4): 244-
264.
[5]
Quan S
un,
Xi
u
q
in
g Ye, W
e
ik
ang
Gu. A N
e
w
Com
b
in
atio
n
Rul
e
of Ev
id
e
n
ce T
heor
y.
Acta Electronica
Sinic
a
. 200
0; 28(8): 117-
11
9.
[6]
Jinh
ai Hu, Z
h
i
guo Yu,
Xush
eng Z
h
a
i
, et al. Resear
ch of Decisi
on F
u
si
o
n
Diag
nos
is of Aero-en
g
in
e
Rotor F
ault b
a
s
ed on Impr
ov
ed D-S T
heor
y.
Acta Aerona
utica et Astron
autica Si
nic
a
. 201
4; 35(2):
436-
443.
[7]
Z
hung
a Li
u, Yo
ngme
i
Ch
en
g, Quan Pa
n, et a
l
. W
e
i
ght Evi
d
e
n
ce C
o
mbi
nati
on for Mu
lti-Se
nsor C
onflic
t
Information.
C
h
in
ese Jo
urna
l of Sensor a
nd
Actuators
. 200
9; 22(3): 36
6-3
70.
[8]
Murph
y
CK. C
o
mbi
n
in
g B
e
li
e
f
F
unctions
w
h
en Ev
ide
n
ce
C
onflicts.
D
e
cisi
on S
u
p
port Sy
stems
. 200
0;
29: 1-9.
[9]
Yong
Den
g
, W
enkan
g Shi,
Z
henfu Z
hu.
Effici
ent Com
b
inati
on Ap
pro
a
c
h of Co
nflict Evide
n
ce.
J.
Infrared Millim
. Waves
. 2004; 23(1): 27-
32.
[10]
Qing Y
e
,
Xi
ao
p
i
ng
W
u
, Ye
xin
Song. Ev
id
enc
e C
o
mbi
nati
on
Method
Bas
e
d
on th
e W
e
ight
Coeffici
ents
and th
e Co
nfli
ction Pro
bab
ilit
y D
i
stributi
on.
System En
gi
n
eeri
ng a
nd El
e
c
tronics
. 20
06;
28(7): 10
14-
101
6.
[11]
Xu
ezh
o
n
g
Yin,
Jiegu
i W
ang.
A Ne
w
Com
b
i
natio
n Ru
le of
Heterog
e
n
e
o
u
s
Sources D
a
ta F
u
sion
on
Conflict Evi
den
ces.
Journa
l of Astronautics
. 2
009; 30(
4): 166
3-16
66.
[12]
W
R
Liu. Anal
yzi
ng the D
egre
e
of Con
fl
ict a
m
ong Be
lief F
unctio
n
s.
Arti
fi
cial Intell
ig
enc
e.
2006; 17
0
:
909-
924.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Weighted E
v
ide
n
ce Com
b
ination Meth
od Base
d on
Im
proved
Co
nflict… (Xiao
c
he
n Xing)
533
[13]
Z
hung
a Liu, Y
ongm
ei Ch
eng
, Q
uan Pan, et al. Combin
ation of
W
e
ig
hted Bel
i
ef F
unctions Bas
e
Evide
n
ce Dista
nce an
d Co
nfli
cting Bel
i
ef.
Co
ntrol T
heory &
Appl
icatio
ns
. 2
009; 26(
12): 14
39-1
442.
[14]
Bo L
i
,
Xins
he
n
g
W
a
n
g
, Kai
x
i
ng Z
h
ou,
et al.
St
ate Mo
nitori
ng
and
F
a
u
l
t Identific
atio
n fo
r Spac
ecraft
und
er Sens
or F
ault.
Journa
l of Astronautics
.
2013; 34(
10): 136
2-13
69.
[15]
Deqi
an
g H
an,
Yi Yan
g
, Ch
on
gzha
o H
an. Ad
vanc
es in DS Evide
n
ce
T
heo
r
y
a
nd Rel
a
ted
Discussi
ons
.
Contro
l and D
e
cision
, 2
0
1
4
; 29(1): 1-11.
[16]
Ying Pe
ng, Z
heng
hui H
u
, Hu
airo
ng She
n
. A Modifie
d
Dista
nce of Evide
n
c
e
.
Journa
l of Electron
ics &
Information T
e
chno
logy
. 2
013
; 35(7): 162
4-1
629.
[17]
Yi Yan
g
, Deq
i
ang H
an, C
h
o
ngzh
ao H
an.
Evide
n
ce
C
o
m
b
in
ation
bas
ed
on Multi-
Criter
ia Ra
nk-L
evel
F
u
sion.
Acta Automa
tica Sinica
. 2012; 3
8
(5): 823-
831.
[18]
Z
hung
a L
i
u, Y
ongm
ei
Che
n
g
,
Q
uan P
an,
et al. T
a
rget
Identific
atio
n b
y
A
d
a
p
tive C
o
mbin
ation
o
f
Conflicti
ng Evi
denc
e.
Acta Aerona
utica et As
tronautic
a Sini
ca
. 2010; 3
1
(7)
:
1426-1
4
3
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.