TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 10, Octobe
r 20
14, pp. 7223
~ 723
2
DOI: 10.115
9
1
/telkomni
ka.
v
12i8.524
7
7223
Re
cei
v
ed
No
vem
ber 2
9
, 2013; Re
vi
sed
Jul
y
2, 2014;
Acce
pted Jul
y
28, 201
4
Path Planning for Coalmine Rescue R
obot Based on
Hybrid Adaptive Artificial Fish Swarm Algorithm
Yao Zheng
-Hua*
1,2
, Ren
Zi-Hui
1
, Zhu Xian-Hu
a
2
, Li Shi-Chun
2
1
School of Infor
m
ation a
nd El
e
c
trical Eng
i
ne
e
r
ing,
Ch
ina U
n
i
v
ersit
y
of Min
i
n
g
and T
e
chno
l
o
g
y
(CUMT
)
,
Chin
a
2
School of Mec
han
ical a
nd El
ectrical En
gin
e
e
rin
g
, Yangtze
Normal U
n
iv
ersit
y
(YZ
NU), C
h
in
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: emscto@qq.
com
A
b
st
r
a
ct
F
o
r the pro
b
le
m w
i
th i
m
pr
eci
s
e opti
m
al so
l
u
tion
and r
e
d
u
c
ed co
nverg
e
n
c
e efficie
n
cy o
f
basic
artificial fish swarm
algori
th
m (
BAF
SA) in the late, the ad
apti
v
e enh
anc
ed p
r
ey beh
avi
o
r of artificial fish a
n
d
the seg
m
e
n
ted
adaptiv
e strategy of artificia
l
fish
’
s
vi
ew
and step w
e
re desi
gne
d. T
h
e
hybrid a
d
a
p
ti
ve
artificial
fish
s
w
arm
alg
o
rith
m (
H
AAF
SA)
w
a
s structured
by th
e
ad
aptiv
e e
n
h
ance
d
pr
ey b
e
h
a
vior
a
n
d
the
seg
m
e
n
ted a
d
aptive strate
gy
of artificial fi
sh
’
s
view
an
d
step, w
h
ich has be
en v
e
ri
fied on r
e
sear
ch.
Accordi
ng to
th
e ch
aracteristic
s of the c
o
a
l
mi
ne resc
ue
env
i
r
on
me
nt, the p
a
th p
l
an
ni
ng
e
n
viro
nment
mo
del
w
a
s establish
e
d
in tw
o-dimen
s
ion
a
l pl
ane a
nd the opti
m
i
z
ation co
nstrain
t
s conditio
n
s w
e
re dispos
ed
by
detectin
g
the
d
i
stance
betw
e
e
n
pat
h secti
o
n
s
and
b
a
rriers.
T
he HAAF
SA
w
a
s app
lie
d to
coal
mine
resc
u
e
robot pat
h pla
nni
ng. Si
mul
a
ti
on resu
lts sho
w
ed that the HAAF
SA could
impr
ove the
p
e
rformanc
e of th
e
opti
m
a
l
path.
Ke
y
w
ords
: AF
SA, path pla
nni
ng, enh
anc
ed
prey,
seg
m
e
n
ted ad
apti
on, re
scue rob
o
t
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Chin
ese co
al
prod
uction
accou
n
ts the
wo
rld’
s for
about 1/3, b
u
t the mine death toll
accou
n
ts for
nearly 80% o
f
the world. In recent
years Chi
na co
al
mine one mill
ion tons mo
rt
ality
has
sh
own d
e
clini
ng tre
n
d
,
and the
situ
ation of
produ
ction
safety h
a
s
b
een im
proved. Com
p
a
r
ed
with oth
e
r m
a
jor coal-pro
duci
ng
co
unt
ries the
ga
p
has still
exist
ed, the
r
efore
the
pro
d
u
c
tion
safety situ
ation i
s
still g
r
i
m
. In ad
ditio
n
to the
po
or co
al
sea
m
condition
s, the
lower deg
re
e of
mining
autom
ation an
d un
even level p
r
actitione
rs
of
mine the
re
aso
n
for
Chi
na coalmi
ne
high
death toll
wh
ich
ca
n n
o
t
be ig
nored i
s
the
outd
a
ted
coalmi
ne
rescu
e
te
chn
o
logy. After t
he
disa
ster, re
scue
wo
rkers can
not
info
rm
ed of the di
saster i
n
form
a
t
ion quickly a
nd a
c
curately,
and
the re
scu
e
staffs with equipm
ents can
rea
c
h
t
he disa
ster
area quickly,
that result
s del
ay on
resc
ue work
[1-3].
Gene
rally p
a
th pla
nnin
g
p
r
oblem
mea
n
s that the
opti
m
al p
a
th i
s
f
ound
on
the
planni
ng
area
from
a
given sta
r
ting
point to the
desti
n
a
tion p
o
int, whi
c
h
can me
et ce
rt
ain pe
rforma
nce
metrics u
nde
r the con
s
trai
nt con
d
ition. Path
planni
n
g
for mine
re
scue robot m
ean
s finding
a
path from
the
startin
g
poi
n
t
to the target
point
un
de
r the envi
r
onm
e
n
t with ob
sta
c
le
s. The
pat
h
sho
u
ld meet
the spe
c
ified
requi
reme
nts, whi
c
h sho
u
ld be safe, reliabl
e and t
i
me-con
sumi
ng
shortest
or least-cost. To
some
extent,
path planning capability refl
ects the intel
ligence level
of
mine rescu
e
rob
o
t. Currently the alg
o
rithm
s
of
p
a
th plan
ning
for ro
bot h
a
ve develo
p
ed in
intelligent an
d bioni
c dire
ction, and a se
ries
of research re
sult
s ha
ve been mad
e
[4-7].
2. Impro
v
ement of
AFSA
AFSA is a kind of swa
r
m intelligen
ce optimizatio
n algorith
m
, which sim
u
lats the
intera
ctive social b
ehavi
o
rs
of fish popul
ati
ons t
o
achi
eve swarm intellig
enc. It is ch
iefly
cha
r
a
c
terrize
d
by o
n
ly co
mpari
ng th
e f
i
tness of
th
e
obje
c
t with
ou
t the spe
c
ific
informatio
n, fast
conve
r
ge
nce spe
ed,
a ce
rtain
ad
apt
ive
cap
a
city of se
arch
spa
c
e
a
nd a
ce
rtain robu
stne
ss
of the
para
m
eters
choice. But the fish swa
r
m
algorithm
h
a
s defe
c
ts too
,
includin
g
im
pre
c
ise optim
a
l
solutio
n
and t
he latter re
du
ced
conve
r
ge
nce effici
en
cy.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 10, Octobe
r 2014: 722
3
– 7232
7224
2.1. Basic AFSA
In a d
-
dime
nsional
sea
r
ch
spa
c
e
there a
r
e
N
piece
s
of artificial fish. T
h
e state
vecto
r
of
artificial fish
positio
n is
expre
s
sed a
s
)
,
,
,
(
2
1
d
x
x
x
X
. The foo
d
con
c
entration of
artificial
fish po
sition i
s
expressed
as
)
(
X
f
Y
, in which
X
is the se
a
r
ch
optimization variabl
e of
artificial fish
state, and Y
is the fitne
s
s fun
c
ti
on. T
he di
stan
ce
betwe
en two
artificial fish
is
defined
a
s
j
i
j
i
X
X
d
,
.
Cro
w
di
ng fa
ctor
repre
s
e
n
ts the
crowd d
egr
e
e
of
fish
swarm.
The
artificial fi
sh’
s
view i
s
rep
r
e
s
sed
as
visual
.
step
rep
r
esents the l
a
rge
s
t moving
step
of artifici
al
fis
h
.
num
try
_
repre
s
e
n
ts the large
s
t number of attempts for prey behavior o
f
artificial fish. The
core id
ea
s of
AFSA mod
e
l
are p
r
ey b
e
havior,
swarm be
havioror, follow
beh
a
v
ior an
d
ran
d
o
m
behavio
r.
2.1.1. Pre
y
Behav
i
or
The
cu
rre
nt state of artifici
al fish i
s
defi
ned a
s
i
X
. Another
state
j
X
is
selected i
n
it
s
view ran
doml
y
, which is ex
pre
s
sed a
s
:
()
Rand
visual
X
X
i
j
()
Rand
is a ran
dom
numbe
r bel
ong to the close
d
interva
l
]
1
,
0
[
. If the state
j
X
is
s
u
pe
r
i
or
to
the
s
t
a
t
e
i
X
, the artific
i
al fish with s
t
ate
i
X
will
move a
step i
n
t
he di
rection of state
j
X
.
()
1
1
Rand
step
X
X
X
X
X
X
t
i
j
t
i
t
i
t
i
If the state
j
X
is not sup
e
ri
or to the state
i
X
, the artifical
fish will
con
t
inue to try to sele
ct
anothe
r ne
w
state
j
X
. The art
i
fical fish
rep
eats to attem
p
t new
state
for
num
try
_
times
.
If i
t
has not been still satisfi
ed with the conditions
of prey, the artifical fish will move a step
rand
omly.
()
*
1
Rand
visual
X
X
t
i
t
i
2.1.2. S
w
a
r
m
Behav
i
or
The
cu
rrent
state of
artifi
cial fi
sh i
s
d
e
fined
as
i
X
. The
artific
i
al
fis
h
adds
up the
numbe
r
f
n
of it
s n
e
igh
borho
od p
a
rtne
rs i
n
the
ra
nge
of
visual
d
j
i
,
, and
find
s the p
a
rtn
e
rs
’
cente
r
C
X
.
f
n
i
i
C
n
X
X
f
1
If
i
f
C
Y
n
X
/
,that mean
s the
r
e i
s
m
o
re fo
od i
n
th
e le
ss
cro
w
d
ed pa
rtne
rs center,
the artifici
al fi
sh
will move
a step towards in t
he
direction of the partne
rs’
center, otherwise it
perfo
rms p
r
e
y
behavior.
()
1
1
Rand
step
X
X
X
X
X
X
t
i
C
t
i
C
i
t
i
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Path Plannin
g
for Coalm
i
n
e
Re
scue Ro
bot Based o
n
Hyb
r
id Ada
p
tive… (Y
ao Zh
eng-Hu
a)
7225
2.1.3. Follo
w
Behav
i
or
The
current
state of a
r
tificial fish is
d
e
fined a
s
i
X
. The artific
i
al fis
h
adds
up the
numbe
r
f
n
of its neig
hbo
rho
od pa
rtners i
n
the ran
ge
of
visual
d
j
i
,
, and finds the partne
r
with state
j
X
, whose fitness functio
n
value
j
Y
is
the bes
t. If
i
f
j
Y
n
Y
/
, it indic
a
tes that the
partne
r
with
t
he state
j
X
has highe
r foo
d
con
c
e
n
tration
and l
e
ss
oth
e
r a
r
tificial fi
sh cro
w
din
g
around it. The artificial fi
sh will move
a step in the direction of
the partner
j
X
, otherwise it
perfo
rms p
r
e
y
behavior.
()
1
1
Rand
step
X
X
X
X
X
X
t
i
j
t
i
j
i
t
i
2.1.4. Rando
m Behav
i
or
The rand
om
behavio
r will
be excuted if the artificial
fish’s fitne
s
s is not imp
r
ov
ed after
the three
kin
d
s b
ehavio
rs above have
been im
ple
m
ent
ed. Th
e
artificial fish
sele
cts a
st
ate
rand
omly in its view, and then move
s a step toward
s
this state. Ra
ndom be
havi
o
r help
s
artifi
cial
fish to escap
e
from local o
p
timal and go
forwa
r
d to the global o
p
timal solutio
n
[8-11].
()
1
Rand
visual
X
X
t
i
t
i
2.2. Impro
v
e
d
Strateg
y
fo
r HAAFSA
Adaptive enh
anced prey pro
c
e
ss i
s
used to im
pro
v
e the artificial fish prey behavio
r
again
s
t the
i
nefficient
pre
y
behavio
r ab
out BAFSA. The a
r
tificial f
i
sh’
s
view an
d ste
p
in BAF
SA
were un
cha
n
ged, so the
algorith
m
co
nverge
nc
e speed redu
ce
d in the later and the opti
m
al
solutio
n
accu
racy
wa
s not
high. A seg
m
ented ad
ap
tive strategy wa
s de
sign
e
d
to improve
the
algorith
m
’s
converg
e
n
c
e
spe
ed an
d o
p
timal accu
ra
cy
by chan
gi
ng the si
ze
of artificial fish’
s
view and
step
.
2.2.1. Adap
tiv
e
Enhance
d
Pre
y
Beha
v
i
or
Artificial fish
sho
u
ld find b
e
tter po
sition
s as m
u
ch a
s
po
ssi
ble in
its view, and move
towards bett
e
r p
o
sitio
n
s
quickly to
re
duce u
nne
ce
ssary
ran
d
o
m
beh
avior.
Whe
n
the
prey
behavio
r coul
d not be
com
p
leted
within
basi
c
try ti
me
s, the a
r
tificia
l
fish p
r
ey be
havior
woul
d
be
cha
nge
d fro
m
the ba
sic prey state t
o
anothe
r
p
r
ey state aut
omatically, which
reali
z
e
d
the
adaptive tran
sform
a
tion of prey beh
avior.
Greate
r
vie
w
, the a
r
tificial
fish
could
find gl
obal
extremum
a
nd
be
conve
r
g
e
n
t more
easily. Th
e g
r
eate
r
of a
r
tificial fish’s
mo
ving st
ep, the
optimization
pro
c
e
s
s converge
d fa
ster.
If
the probl
em
with local extrema wa
s not
very pr
omin
e
n
t, increa
sin
g
the numbe
r of attempt times
coul
d redu
ce
the a
r
tificial
fish
ran
dom
wal
k
, an
d i
m
prove
co
nverge
nce efficiency.
When
the
ca
se with p
r
ominent lo
ca
l extrema was serio
u
s, redu
cing the
attempt times of prey co
uld
increase the
probability of
the artifici
al
fish random
moving,
and
overcome the
impact of l
o
cal
extrema. The
r
efore, the artificial fish a
d
just
e
d
view,
moving step
and times o
f
prey attempt
automatically acco
rding to
the
followi
ng
Equation
(1),
(2)
and
(3
). It was
co
ndu
cive for succe
ss
of artificial fi
sh’
s
p
r
ey an
d the alg
o
rit
h
m pe
rf
orm
a
nce i
m
prove
m
ent. The a
r
tificial fish’
s
prey
behavio
r with
adju
s
ted pa
ramete
r was calle
d ada
ptive enhan
ce
d prey b
eha
vior, whi
c
h
wa
s
helped to improve the probab
ility of prey success [12].
a
visual
enhance
v
_
(
1
)
b
step
enhance
s
_
(2)
c
num
try
enhance
t
_
_
(
3
)
m
,
,
2
,
1
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 10, Octobe
r 2014: 722
3
– 7232
7226
Whe
r
e ,
enhance
v
_
represe
n
ts the view of enh
an
ced p
r
ey.
enhance
s
_
re
pre
s
ent
s
the step of it
s enh
an
ced
p
r
ey.
enhance
t
_
rep
r
e
s
e
n
ts the attem
p
t times of its enhan
ce
d prey.
re
pre
s
e
n
ts
the current
numbe
r
of e
nhan
ce
d p
r
e
y
times.
m
represent
s the maximum
executio
n time of enhan
ce
d prey. Para
meters
a
,
b
and
c
are relate
d to the artificial
fish’s vie
w
and step, wh
ich co
uld be set
a
c
co
rding
to
the
the si
ze of
ba
si
c a
r
tificial
fi
sh’s view,
ste
p
a
n
d
attempt times of prey.
Performi
ng th
e above
pro
c
ess until the
set nu
mbe
r
o
f
attempt time wa
s rea
c
hed
. Whe
n
the numbe
r
of enhan
ce
d
prey time reached
m
, if
the artificial
fish could not still pre
y
su
cc
es
sf
ully
,
t
he si
ze
of
ar
t
i
f
i
cial
fish
's view,
step
and
attempt time
s of
prey
wou
l
d return
to th
e
initial value and execute ra
ndom be
havi
o
r.
2.2.2. Impro
v
ed Vie
w
o
f
Artificial
Fish based on Se
gmente
d Ad
aption
In BAFSA, artificial fish’s size of visi
on an
d ste
p
is fixed. Th
e artifici
al fish’s vie
w
determi
ne
s the ra
nge
of sea
r
ch, an
d its step
d
e
termin
es th
e co
nverg
e
n
c
e
spee
d a
nd
optimizatio
n accuracy. Wh
en the artifici
al fish’s
view i
s
narro
w, its prey and ran
dom beh
avior are
dominate. wh
en its view is
vast, its follow beh
avior is
more p
r
omi
n
ent.
At the initial
stage
of the f
i
sh al
go
rithm, the
va
s
t
view
c
o
u
l
d
in
duc
e
a
r
tific
i
al fis
h
to find
the glo
bal o
p
timal solution,
at the
sam
e
time with
a la
rge
r
size of
step artifici
al fi
sh
co
uld m
o
ve
clo
s
er to the
optimal
sol
u
tion qui
ckly, so it
m
a
kes
the alg
o
rithm
be
co
nverg
ent. Late i
n
t
he
algorith
m
, artificial fish a
ggre
gate
s
around t
he op
timal solutio
n
in a small
area with h
i
gh
probability. If the view i
s
still large now, the ar
tificial fish coul
d ignore the highest food
con
c
e
n
tration
area, thus it will prey ineffi
cient
ly and p
e
rform m
o
re
rand
om beh
a
v
ior. Therefo
r
e,
a se
gmente
d
adaptive
strategy ha
s be
en de
sig
ned
to
improve th
e artificial fi
sh’s
step. Duri
ng
the execution
time of the al
gorithm, the a
r
tificial
fish’
s
f
ileld of view a
nd ste
p
expa
nded first, the
n
they decrea
s
ed ada
ptively with the algo
rithm perfo
rm
ed.
V
V
l
time
iter
k
visual
adap
visual
^
_
/
*
_
adap
visual
_
rep
r
e
s
ent
s t
he imp
r
oved
artificial
fi
sh’
s
vision
by ad
aptive strateg
y
.
V
k
and
V
l
are p
a
rameters of a
daptive strategy.
time
ite
r
_
is the
curre
n
t iterations. Fi
rst
the
improve
d
artificial fish’
s
vie
w
increa
se
d at the
beginn
ing of the alg
o
rithm, whi
c
h
is benifici
al for
artificial fish to find the glo
bal optimal
solution.
As th
e algo
rithm iteration
s
in
cre
a
sei
ng, artificial
fish’s vie
w
re
duced a
dapti
v
ely. Along with its wi
e
w
re
duced a
dapti
v
ely, the execution
pro
bab
ility
of artificial fish prey be
hav
ior an
d ra
nd
om beh
av
ior
increa
sed,
which i
s
in favor of enh
an
ci
ng
local
sea
r
ch and imp
r
ovin
g accuracy of
optimal soluti
on.
2.2.3. Impro
v
ed Step of
Artificial
Fish based on Se
gmente
d Ad
aption
The larger
step si
ze, the vaster artifici
al
fish’s mov
e
ment range will be, whi
c
h was
con
d
u
c
ive to be co
nverg
e
n
t
as soo
n
as
possibl
e. Lat
er in the algo
rithm, the arti
ficial fish coul
d
not only mi
ss
the glob
al opt
imal solution
easily
with
to
o large
step
size, o
r
get
a solution
with lo
w
accuracy, but
also be
prone to oscillate
back and
fort
h near the optimal so
lution.
That was why
i
t
wa
s difficult t
o
app
roximat
e
the optimal
solutio
n
a
c
c
u
rat
e
ly
.
A
sm
all st
ep
si
ze
wa
s in f
a
v
o
r
of
local sea
r
ch for artificial fish, but the spe
ed of sea
r
chi
ng global o
p
timal solutio
n
wa
s slo
w
e
r
, and
the algorithm was
eas
y
to f
a
ll into loc
a
l optimum.
Early in the algorithm, artifi
cial fish could
enhan
ce the
global se
arch ability with a large
r
step
size, wh
ich m
ade th
e
artificial fi
sh
move cl
ose
r
t
o
a bette
r sol
u
tion an
d ag
greg
ate a
r
ou
nd
the optimal solution a
s
qui
ckly a
s
po
ssi
ble. With
the algorith
m
exe
c
uting,
the d
e
crea
se of st
ep
size hel
ped t
he a
r
tificial fi
sh to
agg
reg
a
te around
th
e optimal
sol
u
tion, re
du
ce
the proba
bility of
over the opti
m
al solutio
n
, and en
han
ce
the ability of
algorith
m
local sea
r
ch in the latter [13-14].
)
^
_
/
(
*
_
S
S
l
time
iter
k
step
adap
step
(
4
)
adap
step
_
rep
r
e
s
ent
s t
he imp
r
oved
artificial fi
sh’
s
step ba
sed on
ad
aptive strategy.
S
k
and
S
l
are para
m
eters of
ada
ptive strategy
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Path Plannin
g
for Coalm
i
n
e
Re
scue Ro
bot Based o
n
Hyb
r
id Ada
p
tive… (Y
ao Zh
eng-Hu
a)
7227
Acco
rdi
ng to
Equation
(4),
first the
si
ze
of
improved
step i
n
crea
se
d at the
be
gi
nning
of
the iteration,
whi
c
h ma
de
the artificial
fish ag
gre
g
a
te to the o
p
timal sol
u
tio
n
fast. With
the
iteration
s
increasi
ng, attenuati
on streng
thened, and t
he step si
ze
has de
crea
se
d. At
the same
time the a
dap
tive attenuate
d
ste
p
coo
perated
wi
th the
adaptive
attenuated
view to en
han
ce
th
e
local
sea
r
ch, whi
c
h imp
r
ov
ed the accu
ra
cy of solution
s.
Algorithm p
a
rameters imp
a
ct on the p
e
rfor
m
a
n
c
e o
f
the algorith
m
greatly. When the
attenuation i
ndex is too l
a
rge, it
will
cause the alg
o
rithm to ma
ture ea
rly, and fall into local
optimal soluti
on, even fail
to conve
r
ge
. Acco
rdin
g to the expe
ri
mental stu
d
ie
s, gen
erally t
he
sele
ction int
e
rval of parameters
V
k
and
S
k
was
s
e
t as [1.5,2.5].
S
l
selection interval was
gene
rally
set
as [0.2,0.7].
V
l
sel
e
ctio
n int
e
rval i
s
g
ene
rally set a
s
[
0
.8,3]. The
s
e
paramete
r
s
use
d
for i
m
p
r
oveme
n
t we
re
sele
cted i
n
their
intervals a
c
cordin
g to the
spe
c
ific o
p
timiza
tion
probl
em.
On the surfa
c
e, the increa
ses of a
r
tificial
fish’s vie
w
an
d step in e
n
h
anced p
r
ey b
ehavior
wa
s
contradi
ctory with th
e
adaptive
attenuation
of th
e
m
in al
go
rith
m excutio
n
p
r
oce
s
s, but it
wa
s
not in fact.
When the
ba
sic prey b
ehavi
o
r
could
not be
a
c
hieve
d
within
the nu
mber of
attempt
times, enha
n
c
ed p
r
ey beh
avior wo
uld b
e
execute
d
.
The ada
ptive improvem
ents of artificial fish’s
view an
d ste
p
wa
s reali
z
ed by imp
r
ov
ing t
he
vie
w
and step
of basi
c
AFSA with
segme
n
ted
adaptive st
ra
tegy. The art
i
ficial fish’
s
view
an
d ste
p
whi
c
h ha
d
been im
prov
ed by ada
ptive
segm
ented
strategy
were
amplified
d
u
ring
th
e
pe
riod
of e
nha
nce
d
p
r
ey b
ehavior excu
ted.
Enhan
ced
prey behavio
r
ran throug
h
t
he
e
n
tire alg
o
rithm excuti
on
p
r
o
c
e
ss, essentially which
wa
s the imp
r
oved meth
o
d
for artifici
a
l
fish
prey behavior i
n
the entir
e algo
rithm
exe
c
ution
pro
c
e
ss. T
he
segm
ented
a
daptive impro
v
ement for a
r
ti
ficial fish’s vi
ew a
nd ste
p
wa
s excuted in
each iteration
in the algorit
hm pro
c
e
s
s.
2.3. Algorith
m
Verificatio
n
The functio
n
1
F
and
2
F
are took for example
to prov
e the validation of the HAAFSA
respec
tively. Func
tion
1
F
has
a si
ngle
ma
ximum at po
int (0,0
), an
d some l
o
ca
l extremum
s
spread
aroun
d the extrem
e point. Fu
n
c
tion
2
F
is a t
y
pical o
p
timization
pro
b
le
m with mo
re
extremum,
which
obtai
ns
the glo
bal
op
timum value
3600
at p
o
int
(0,0
). Several lo
cal
mini
ma
points l
o
cate
d on (-5.1
2
,5
.12), (-5.12, -5.12), (5
.12,
-5.12
)
an
d (5
.12,5.12)
sca
tter aro
und t
h
e
global o
p
timu
m, which o
b
tain the local extrema 27
48
.7823.
Paramete
r selectio
n: total numbe
r of artificial fish
80
N
, maximum number of
iteration
s
50
max_
gen
, moving
ste
p
5
.
0
step
, view
2
visual
, ti
mes
of attempt
10
_
mun
try
, the co
nge
stion facto
r
618
.
0
, adaptive pa
ra
meters of e
n
han
ced
pre
y
behavio
r (
5
m
,
1
a
,
2
.
0
b
,
2
c
), adaptive parameters
of view an
d step (
2
S
k
,
2
V
k
,
4
.
0
S
l
,
4
.
2
V
l
). Simul
a
tio
n
conditio
n
s:
CPU Intel
Core i3
-2
33
0M 2.2
G
Hz,
RAM
2G,
operating system Wind
o
w
s7,
simul
a
tion
so
ftware
Matlab_
R2
012a. T
able
1 sh
ows t
he
perfo
rman
ce
comp
ari
s
o
n
a
bout HAAFS
A
and BAFSA for 25 times simul
a
tion
s.
y
y
x
x
y
x
F
)
sin(
)
sin(
)
,
(
1
,
)
10
,
10
(
y
x
2
2
2
2
2
2
2
)
(
)
)
(
05
.
0
3
(
)
,
(
y
x
y
x
y
x
F
,
)
12
.
5
,
12
.
5
(
y
x
Table 1. The
Perform
a
n
c
e
of HAAFSA and BAFSA
Fuction
Algorithm Optimum
Solutio
n
Time
（）
s
Worst Solution
Time
（）
s
Average
F
1
BAFSA 0.99999
3.50
0.99909
3.37
0.99968
HAAFSA 1.00000
2.61
1.00000
3.28
1.00000
F
2
BAFSA 3599.36394
3.94
2748.77907
4.49
3405.74121
HAAFSA 3600.00000
3.31
2748.77907
4.31
3542.12176
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 10, Octobe
r 2014: 722
3
– 7232
7228
Table 1 sh
ows the HAAFSA is supe
rior
to the BAFSA. Although the HAAFSA also wa
s
likely to fall into local opti
m
al, the prob
ability had re
duced g
r
eatly
, and the ave
r
age
had
bee
n
improve
d
ob
viously, whe
n
the pro
b
le
m of optim
izi
ng obje
c
t local optimal was serio
u
s.
The
HAAFSA had
stren
g
thene
d the local a
r
ea se
arch,
a
nd improved
the accuracy
of optimum a
n
d
conve
r
ge
nce spe
ed.
3. Path Planning En
v
i
ronment Model
Acco
rdi
ng to
the enviro
n
mental info
rmat
ion, the
area of pl
annin
g
missi
on wa
s
determi
ned,
and the
ma
p
model
was
establi
s
h
ed t
o
plan
the
walk
path fo
r robot effe
ctively.
Obviou
sly, the pla
nnin
g
a
r
ea
wa
s a
two
-
dime
nsi
onal
model fo
r
mo
vement, and
t
he
robot
moti
on
environ
ment
co
uld b
e
expre
s
sed
b
y
two-di
men
s
ion
a
l coo
r
di
nate. Robot
moved
on
two-
dimen
s
ion
a
l finite spa
c
e,
and a
ce
rtai
n numb
e
r
of
impassa
ble area
s were distrib
u
ted
in
the
scope of this
area
s, su
ch a
s
ob
stacl
e
s a
nd dan
gerou
s points.
3.1. En
v
i
ron
m
ent Model
In the plan
ni
ng a
r
ea, the i
m
passa
ble a
r
eas
we
re exp
r
esse
nd
by
convex polygo
n
s a
nd
roun
ds. The site
of re
scu
e
ro
bot
b
egin
n
ing re
scue missi
on wa
s the
sta
r
ting p
o
int
of
the p
a
th
planin
g
. The
end p
o
int of
the path
plan
ing was site
d
as th
e p
o
siti
on
whe
r
e th
e
re
scue
rob
o
t
bega
n rescu
e
op
eratio
ns
throug
h the
o
b
sta
c
le
ar
e
a
. In coordinat
e sy
stem of t
he robot
mov
i
ng
environ
ment
model, the st
arting p
o
int was set at
)
,
(
S
S
y
x
S
, and the target e
ndpoi
nt wa
s set at
)
,
(
T
T
y
x
T
. According t
o
the start a
nd end p
o
siti
ons, the two
-
dimen
s
ion
a
l coo
r
din
a
te sy
stem
wa
s e
s
tabli
s
hed, which
wa
s the e
n
tire are
a
of
re
scue
robot
p
a
th plan
ning.
The e
n
viro
men
t
modle of path
plannin
g
is shown in Figu
re 1.
Figure 1. Environm
ent Mod
e
l
3.2. Path Re
presen
ta
tion
As it is sho
w
n
in Figure 1, p
a
th planni
ng i
s
to find a set
of points:
T
p
p
p
S
P
n
,
,
,
,
,
1
-
2
1
In the glob
al
spa
c
e,
whi
c
h
are
con
n
e
c
te
d adja
c
e
n
tly without o
b
sta
c
le
s an
d da
n
gero
u
s
point
s so
that the p
a
th
from
sta
r
t p
o
int to the
ta
rget
point l
e
n
g
th ha
s
bee
n
plan
ned.
1
n
parallel
lines
were marke
d
out parallel to the Y ax
is, which divid
e
d
the X axis into
n
section
s
between the
start an
d en
d
point. On ea
ch pa
rllel lin
e
a point was
sele
cted a
s
t
he refe
ren
c
e
point, from th
e
starting p
o
int S through the
refere
nce poi
nts to t
he targ
et point T . A
path wa
s obt
ained, nam
el
y:
T
y
x
y
x
y
x
y
x
S
Path
n
n
i
i
,
,
,
,
,
,
,
,
,
,
1
1
2
2
1
1
，
0
20
40
60
80
100
0
10
20
30
40
50
60
70
80
90
10
0
X a
x
is
Y axi
s
E
n
v
i
r
o
n
m
en
t
M
o
del
123
…
…
n
-
1
n-
2
S
P1
P2
P3
P4
P5
P6
P7
P8
T
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Path Plannin
g
for Coalm
i
n
e
Re
scue Ro
bot Based o
n
Hyb
r
id Ada
p
tive… (Y
ao Zh
eng-Hu
a)
7229
The d
o
t
i
i
y
x
,
r
e
p
r
e
s
en
te
d
th
e re
fe
r
e
nc
e p
o
i
nt c
o
or
d
i
na
te
s o
n
th
e
i-
th
pa
r
a
lle
l.
R
o
bo
t s
t
a
r
ted
from the sta
r
ting point, and
moved alon
g
the referen
c
e
point until the end [15-16]
.
4. Path Planning Bas
e
d on Impro
v
ed AFSA
R
e
fe
re
nc
e
po
in
ts
w
e
r
e
on
th
e
1
n
parall
e
l lines, whi
c
h were perpe
ndicular
to th
e
X
axis and divi
ded it into
n
section
s
be
tween the st
art and
end
points. The
r
efore, durin
g
executio
n of t
he alg
o
rithm,
only the val
ue of ea
ch
re
feren
c
e
point’
s
verti
c
al axi
s
was adju
s
t
ed,
and the
ab
scissa value
was u
n
chan
ge
d. Thu
s
the p
a
th co
uld b
e
indicated o
n
l
y
by the vertical
coo
r
din
a
tes o
f
referen
c
e p
o
ints. Each
state of artificial fish rep
r
e
s
ented a path.
1
2
1
,
,
,
n
i
x
x
x
X
Parameter
i
x
was
value of the i-th
referenc
e
ordinate,
i =
1, 2, ... , n-1. Artific
i
al fis
h
ac
ted
rand
omly one
time along wi
th its
state ch
ange
d, that produ
ce
d a ne
w path. If the
fitness valu
e of
the new p
a
th wa
s su
peri
o
r
to the original
one,
the artificial fish
woul
d update its
state.
4.1. Fitness
Ev
aluation Function
Whe
n
coal mine
di
sa
sters
h
apen
ded, rescu
e
robot
s mu
st
rea
c
h
the poi
nt of
disa
ste
r
operation
s
a
s
quickly a
s
p
o
ssible, the
r
e
f
ore the p
a
th
length
wa
s th
e main e
s
tim
a
ting criteri
o
n
s
.
The total
pat
h dista
n
ce
could b
e
exp
r
essed
as
the
sum
m
ation
distan
ce
of e
a
ch
path
se
ction
from the sta
r
t point to the end with the re
feren
c
e poi
nts, namely:
2
1
2
1
2
1
2
1
2
1
2
1
2
1
n
T
n
T
n
i
i
i
i
i
S
S
y
y
x
x
y
y
x
x
y
y
x
x
Y
n
x
x
i
x
x
S
T
S
i
/
)
(
*
Whe
r
e,
)
,
(
S
S
y
x
is t
he p
a
th
sta
r
ting p
o
int
coo
r
din
a
tes,
and
)
,
(
T
T
y
x
is th
e
end
coo
r
din
a
tes.
The i-th
dim
ensi
on
state
value of art
i
ficial fish i
s
i
y
. The i-th
ref
e
ren
c
e
point
c
o
or
d
i
na
te
is
)
,
(
i
i
y
x
, i =
1, 2, ..
. ,
n-1.
4.2. Constrai
nt Processi
ng and Collision Detection
As the robot
path is
determi
ned by the
path sectio
n
,
detec
ting whether
it colli
ded with
the ob
stacl
e
s, what shoul
d
be carried
o
u
t se
ct
ionally.
S
een by the
environ
ment
model, the p
a
th
planni
ng co
n
s
traint i
s
expressed a
s
:
k
D
D
D
D
D
2
1
Whe
r
e,
i
D
rep
r
e
s
ente
d
the i
n
feasibl
e
a
r
ea
s
k
i
,
,
2
,
1
. Acco
rding
to the expression
of the
path if the path can not ha
ve any intersection with
th
e obsta
cle re
gion, the con
s
traint
s wo
ul
d be
sat
i
sf
ie
d.
4.2.1. Proces
sing of Co
ns
traint
The two
asp
e
cts
of the p
a
th plan
ning
con
s
trai
nt pro
b
lem shoul
d
be con
s
ide
r
e
d
, one i
s
the processin
g
of va
riable
boun
dar
y, a
n
d
the
other i
s
the avoi
dan
ce of the
robot
ob
stacl
e
. Th
e
artificial fi
sh
state va
riable
s
m
a
y cro
s
s
the bo
und
ary
with th
e a
r
tificial fi
sh
migrating d
u
rin
g
t
he
time of algo
rithm execution
.
After the art
i
ficial
fish
acti
on, the state
variable
s
of e
a
ch
dimen
s
io
n
woul
d b
e
ch
ecked, if th
e
variabl
es ex
cee
d
th
e
bo
rder, whi
c
h would be set as
th
e
valu
e
of
boun
dary. Thi
s
mea
s
u
r
e m
ade the artificial fish sea
r
ch in a given range, an
d the new lo
catio
n
is
also b
enefi
c
ia
l for artificial fish to find a n
e
w better p
o
sition.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 10, Octobe
r 2014: 722
3
– 7232
7230
min
min
max
max
,
x
x
x
x
x
x
x
i
i
i
，
Und
e
r th
e
co
ndition
of en
suring
the
artifi
cial fi
sh
state
value i
n
the
valid ra
nge, it
sh
ould
be co
nsi
dere
d
wheth
e
r th
e path overl
aped
with
th
e obsta
cle
s
.
Duri
ng the e
x
ecution of t
h
e
algorith
m
, the feasible
path
may beco
m
e
infeasibl
e
,
after the artifici
al fish a
c
ted
once time. For
any artifici
al f
i
sh
)
,
,
,
,
(
1
2
1
n
n
x
x
x
x
X
, if any co
nne
ction p
a
th
between t
w
o
adja
c
ent
poi
nts
coin
cid
ed
with the ba
rri
ers, the
state
of artificial fi
sh m
u
st
b
e
update
d
until
the co
nditio
n
is
sat
i
sf
ie
d.
4.2.2. Dete
cti
on of Poly
go
nal Obstacle
For polyg
on diso
rde
r
s, first the position
a
l
re
lation
b
e
twee
n the each vertex abscissa of
polygon an
d the abscissa
of segmented path
two endp
oints sh
ould be
det
ermin
a
ted. The
absci
ssa of p
o
lygon vertex
which
is o
u
tside the ab
scissa of path e
ndpoi
nts, ha
s no effect on t
h
e
path. The
pol
ygon vertexe
s
bet
wee
n
th
e path t
w
o
e
ndpoi
nts
whi
c
h dist
ributed
of the same
side
of the path
se
gment, ha
s n
o
effect on th
e path too.
O
b
viously, whe
n
the vertexe
s
of the p
o
lyg
on
distrib
u
te the
same si
de
of the path, it indi
cate
s that the seg
m
ent path o
v
erlape
d with
th
e
inacce
ssible
area
s,
whi
c
h
is i
n
fea
s
ible
. The
coo
r
di
nate of th
e referen
c
e
po
sition should
be
adju
s
ted.
4.2.3. Dete
cti
on of Circula
r
Obstacle
Wheth
e
r pat
h segme
n
ts overla
p
wit
h
the
ci
rcul
ar
ob
stacle,
the
ca
se
should
be
con
s
id
ere
d
from two a
s
p
e
c
ts
se
pa
ratel
y
, one i
s
th
e
ce
nter
coo
r
d
i
nate of
the
circle
o
b
sta
c
le
s
betwe
en the
path end
point
s,
the other i
s
outside.
Let the i
-
th center
co
ordi
n
a
te of the
ba
rrie
r
regio
n
a
s
Ri
Ri
y
x
,
, and the j
-
t
h
refe
ren
c
e
point
j
j
y
x
,
meet:
i
Ri
j
Ri
j
R
y
y
x
x
2
2
Thus
refe
ren
c
e poi
nts ma
y be kept out
side the
obst
a
cle
s
region.
i
R
represents th
e radi
us of th
e
i-th ci
rcular b
a
rri
er. A safe
distan
ce
△
i
s
set to en
sure th
at there is
som
e
secu
rity distan
ce
betwe
en the
rob
o
t walki
n
g path
and
the b
a
rriers, f
o
r the
ro
bot
is a
b
stracte
d
into a
moving
particl
e.
Whe
n
the
ref
e
ren
c
e
poi
nts are
out
side t
he thr
eateni
n
g
re
gion, th
e
path may
still
overla
ped
with the inaccessible areas.
At
this mom
ent a vertical
line is d
r
awe
d
from the ce
nter
Ri
Ri
y
x
,
of
each ina
c
cessible a
r
ea to t
he path secti
on, and the p
edal ha
s be
e
n
got as
dj
dj
y
x
,
. When the
pedal
is out
side the
path
se
ction, thi
s
path
se
ctio
n
is fea
s
ibl
e
, o
t
herwi
se
it i
s
ne
ce
ssary t
o
determi
ne th
e size of dist
ance betwee
n
the the ped
al
dj
dj
y
x
,
and the ce
nter
Ri
Ri
y
x
,
of an
y
inacce
ssible
area
s. Wh
en the
di
stan
ce
is
lo
nge
r
th
an
the
radi
us of
the i
n
a
c
cessible a
r
ea
s, th
e
path se
ction i
s
feasi
b
le, otherwise the refren
ce
poi
nts of this path
must be adj
u
s
ted [15, 17].
i
Ri
dj
Ri
dj
R
y
y
x
x
2
2
4.3. Algorith
m
Steps
1) Th
e pla
n
n
i
ng re
gion
al
environ
menta
l
dat
as
are i
m
porte
d to g
enerate the
model of
planni
ng area
.
2) Initializi
ng
the paramete
r
s
of artificial
fi
sh (i
nclu
din
g
the num
be
r of artifici
al fish
N
,
artific
i
al fis
h
v
i
ew
visual
, moving step
step
, maximu
m numb
e
r
of iteration
s
IT
, tim
e
s
of attempt
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Path Plannin
g
for Coalm
i
n
e
Re
scue Ro
bot Based o
n
Hyb
r
id Ada
p
tive… (Y
ao Zh
eng-Hu
a)
7231
num
try
_
, c
r
ow
d
i
ng
fa
c
t
o
r
), gen
erating
N
piec
es
of artific
i
al fis
h
, formatting the initial
artific
i
al fis
h
swarm.
3) The
curre
n
t
iteration is set as
0
ITtimes
, and the param
eters of adaptive e
nhan
ce
d
prey and a
d
a
p
tive strategy
are initialized
.
4) Starting the algo
rithm,
and prey b
ehav
ior, swa
r
m beh
avior,
fellow beha
vior and
rand
om be
ha
vior are
perf
o
rme
d
by ea
ch a
r
tificial
fish then th
e fitness functio
n
value of these
behavio
rs will
be
com
p
a
r
ed
with
ea
ch
other. T
he
beh
avior
with o
p
timal fitness fu
nction
value
will
be sel
e
cte
d
to perfo
rm.
5) T
he vie
w
and
step
of
artificial
fish
have b
een
m
odified
ada
ptively, and d
e
termin
ed
wheth
e
r to m
eet the call to
strengt
hen th
e
prey beh
avior of the ada
ptive process;
6) After each
artificial fish
has a
c
ted on
e time
, comp
ared its fitne
s
s to the bulle
tin board, if
the fitness i
s
sup
e
rio
r
to bu
lletin board, the bulletin b
o
a
rd will b
e
up
dated.
7) Determi
n
in
g wheth
e
r
ITtmes
ha
s re
ached th
e maximum
numbe
r of ite
r
ation
s
IT
, i
f
the maximu
m numb
e
r of
iteration
s
i
s
rea
c
he
d, the
optimal p
a
th
will b
e
outp
u
t and
algo
rithm
end
s, otherwi
se
1
ITtmes
ITtmes
, and go to step 4.
5. Simulation
The are
a
of robot path pla
nning is
set as
100
100
in the coordinate syste
m
, and the
point S
)
10
,
10
(
and T
)
100
,
100
(
are set as the starting a
n
d
end point
s. The numbe
r of artificial
fis
h
is
set as
20
N
. The maximum numb
e
r
of iterations i
s
set a
s
200
ITtimes
. The view
and
step
of a
r
tificial fish a
r
e set
as
20
visual
, moving ste
p
5
step
. Prey attempt time is
s
e
t
as
20
_
mum
try
. The cong
estion fa
ctor i
s
set a
s
618
.
o
.
Paramete
rs
of adaptive enhan
ce
d pre
y
are set as (
5
m
,
1
a
,
5
.
0
b
,
2
c
). The
para
m
eters
of segm
ente
d
adaptiv
e strategy are
set as (
2
S
k
,
2
V
k
,
4
.
0
S
l
,
4
.
1
V
l
).
Simulation co
ndition
s are t
he sam
e
as
above. Fi
gu
re 2 sho
w
s the optimal pat
h diagram
s of th
e
BAFSA and the HAAFSA, whi
c
h have b
een exe
c
uted
for 20 times, respe
c
tively.
(a) BAFSA
(b) HAAFSA
Figure 2. Path Comp
ari
s
o
n
Figure 2 sh
o
w
s the
HAAFSA has stre
n
g
thene
d t
he local
sea
r
ch p
a
rticul
arly, which h
a
s
gotten the s
h
orter path. Table 2 shows the
results
,
after the BAFSA and HAAFSA have been
execute
d
for
20 times sep
a
rately. Acco
rding to
T
able
2, althoug
h the spen
ding t
i
me of HAAF
SA
wa
s a little longer tha
n
BAFSA, the path of HAAFSA is mu
ch short
e
r than BAFS
A
greatly.
0
20
40
60
80
100
0
10
20
30
40
50
60
70
80
90
10
0
X a
x
i
s
Y axi
s
S
T
0
20
40
60
80
10
0
0
10
20
30
40
50
60
70
80
90
10
0
X a
x
i
s
Y a
x
i
s
S
T
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 10, Octobe
r 2014: 722
3
– 7232
7232
Table 2. The
Perform
a
n
c
e
of BAFSA and HAAFSA for Path Planni
ng
Algorithm Mim
Max
Average
Time/s
BAFSA 137.3653
145.6743
141.3926
25.69
HAAFSA 133.2506
139.9176
136.0129
22.05
6. Conclusio
n
The a
daptive
enh
anced
p
r
ey be
havior
has bee
n u
s
ed to im
prov
e artifici
al fish’s
prey
pro
c
e
ss, an
d
the segme
n
ted adaptiv
e
strategy ha
s
been de
sig
n
e
d
to transfo
rm artificial fish’s
view and
ste
p
. The ada
ptive enhan
ce
d
prey pro
c
e
s
s and the
se
gmented
ada
ptive strategy
of
artificial fish’s view and
ste
p
had be
en
combinate
d
to
con
s
tru
c
t the HAAFSA, whi
c
h ha
d be
en
verified. The
HAAFSA wa
s applied o
n
the mine
rescue ro
bot path
plannin
g
, that expande
d the
appli
c
ation fi
elds of AFSA. Acco
rdi
ng t
o
the e
n
viro
n
m
ental
cha
r
a
c
teri
stics of t
he min
e
rescue
robot p
a
th pl
annin
g
area, the re
scue p
a
th planni
ng
model h
ad b
een cre
a
ted. The artifici
al fish
wa
s e
n
code
d
by on
e-di
m
ensi
onal
pa
rameters i
n
st
ead
of two
-
di
mensi
onal
co
ordin
a
tes of
th
e
referen
c
e poi
nts, and th
e
efficien
cy of algorith
m
was imp
r
oved
by this met
hod. In order to
meeting th
e
co
nst
r
aint
con
d
ition, e
a
c
h
se
gmente
d
path
ha
d
bee
n d
e
tected in th
e t
w
o-
dimen
s
ion
a
l
plane. If th
e
con
s
trai
nt
co
ndition
wa
s
n
o
t met, the
re
fence
poi
nt of
this segme
n
ted
path would
b
e
adju
s
ted,
u
n
til it wa
s fea
s
ible.
Whe
n
each
segm
en
ted
path did not
coinci
de with
the infeasi
b
le
area
s, the path se
ction was effectiv
e. Obviou
sly, the result of simulation indi
cate
d
that the path plann
ed by HAAFSA was superi
o
r to the
BAFSA.
Referen
ces
[1]
Sun Ji-
p
in
g. R
e
searc
h
o
n
co
al-min
e
safe
pr
oducti
on c
onc
e
p
tion.
J
ourn
a
l of
Chi
na Co
al Society
. 2
011
;
36(2): 31
3-3
1
6
.
[2]
Sun J
i
-pi
ng. N
e
t
w
o
r
ki
ng tec
h
nol
og
y for
saf
e
t
y
s
u
p
e
rvisi
o
n
s
y
stem
in
a
c
oal
min
e
.
Jo
urnal
of C
h
in
a
Coal S
o
ciety
. 2
009; 34(
11): 15
46-1
549.
[3]
Sun Ji-p
in
g. Safet
y
pr
od
uctio
n
monit
o
rin
g
a
nd commu
nic
a
tion tech
nol
og
y i
n
coa
l
min
e
.
Journa
l of
Chin
a Co
al So
ciety
. 2010; 3
5
(
11): 192
5-1
9
2
9
.
[4]
Mao Ya
ng, C
h
un-zh
e Li. P
a
th Pla
n
in
g a
nd
T
r
acking for Multi-rob
o
t S
y
ste
m
Based
on I
m
prove
d
PSO
Algorit
hm.
Internatio
nal C
onf
erenc
e on Me
chatron
i
c Sc
ie
nce, Electric
Engi
neer
in
g a
nd Co
mputer
.
Jilin. 2
011; 1
6
6
7
-17
70.
[5]
MA Qianzhi, L
E
I Xiu
j
ua
n. Ap
plicati
on of imp
r
oved
p
a
rticle s
w
a
rm optimiz
at
ion
a
l
gor
ithm in
robotic pat
h
pla
nni
ng.
Co
mputer Eng
i
n
eeri
ng an
d App
lica
t
ions
. 201
1; 47
(25): 241-
24
2.
[6]
Z
H
ANG Xi
ao-
yan, Z
H
OU
Xia
o
-
y
ua
n, W
E
I Juan. Glo
b
a
l
p
a
th pl
an
nin
g
f
o
r coa
l
min
e
r
e
scue r
o
b
o
ts.
Journ
a
l of Xi
’
a
n Univ
ersity of Scienc
e an
d T
e
chn
o
lo
gy
. 200
8; 28(2): 32
3-3
26.
[7]
GAO Yang, SUN Shu-d
o
n
g
, HUANG W
e
i-feng. Rap
i
d p
a
th pla
nni
ng o
f
mobile rob
o
t
s
in unkn
o
w
n
envir
onme
n
t.
Applic
atio
n Res
earch of Co
mp
uters
. 2009; 2
6
(
7): 2624-
26
27
.
[8]
Li Xi
ao
lei.
A
Ne
w
I
n
tell
ige
n
t
Optimization Meth
o
d
-Artific
ial Fis
h
S
w
a
r
m Alg
o
rithm.
PhD T
hesis.
Han
g
zho
u
: Z
hejia
ng U
n
ivers
i
t
y
. 200
3: 33-3
8
.
[9]
LI Xiao-lei, SHAO Zhi-jiang,
QIA
N Ji-xin. A
n
Optimizi
ng
M
e
thod
Bas
ed
o
n
Auto
nom
ous
Animats: F
i
sh-
s
w
arm Algorithm.
Systems En
gin
eeri
ng-T
h
eo
ry & Practice
. 2002; (11): 3
3
-3
4.
[10]
Jian
g Min
y
a
n
, Yuan D
o
n
g
fen
g
. Artificial F
i
sh S
w
a
rm Alg
o
r
i
thms and It Applic
atio
ns. Bei
jing: Sci
enc
e
Press. 2012: 4
2
-47.
[11]
Ming
ya
n J
i
a
n
g
,
Don
g
fen
g
Y
u
an, Yo
ngm
ing
Che
ng.
I
m
pr
ov
ed Artific
i
al
F
i
s
h
Sw
ar
m Al
go
rithm
. 20
09
F
i
fth Internatio
nal C
onfere
n
ce
on Na
tura
l Co
mputatio
n. 200
9, 281-2
85.
[12]
ZHANG Yan, CHU
Xi
ao-L
i
. Advanc
ed Artifici
al Fish S
w
a
rm Algorit
hm.
Co
mp
uter Syste
m
Applic
atio
ns
.
201
1; 20(5): 19
9-20
1.
[13]
Jian
g MY, Nik
os E. Mastorak
is, et al.
Multi-t
h
resh
old
i
m
a
g
e
seg
m
entati
o
n w
i
th i
m
prov
e
d
artifici
al fish
sw
arm alg
o
rith
m.
Proce
edi
ng
s of Europe
an
Comp
uting
C
o
nferenc
e (ECC
200
7). 200
7; 117-1
20.
[14]
W
enJie
T
i
an, JiChe
n
g
Li
u.
An Improv
ed
Artificial F
i
s
h
S
w
a
rm Al
gor
ithm for M
u
lti
Rob
o
t T
a
s
k
Sched
uli
ng. F
i
fth Internatio
nal
Confere
n
ce
o
n
Natura
l Com
putatio
n. 200
9; 127-1
30.
[15]
Lei
Xi
uju
an. Sw
a
rm Intelli
ge
n
t
Optimization Algor
it
hms and
T
heir Applic
ati
ons. Beiji
ng: S
c
ienc
e Press.
201
2: 249-
269.
[16]
F
e
i L
i
u, Sh
an
Lia
ng,
Xi
ao
do
n
g
Xi
an. Optim
a
l Path
Pla
n
n
i
n
g
for M
obi
le
R
obot
Usin
g T
a
il
ored
Gen
e
tic
Algorit
hm.
T
E
LKOMNIKA Indones
ian J
ourn
a
l of Electrica
l
Engi
neer
in
g
. 2014; 12(
1): 1-9
.
[17]
LIU Gan
g
, LIU
Yu-bi
n
, Z
H
AO Jie, et a
l
. Pat
h
pl
an
nin
g
for
a ne
w
mi
ne r
e
scue ro
bot b
a
s
e
on
visu
al
tange
nt gr
aphs
.
Journ
a
l
of Jil
i
n
Un
iversity (E
ngi
neer
in
g a
n
d
T
e
chn
o
lo
gy E
d
itio
n).
20
11;
4
1
(4): 1
107-
111
2.
Evaluation Warning : The document was created with Spire.PDF for Python.