TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 10, Octobe
r 20
14, pp. 7478
~ 748
5
DOI: 10.115
9
1
/telkomni
ka.
v
12i8.558
7
7478
Re
cei
v
ed
Jan
uary 6, 2014;
Re
vised Aug
u
st 2, 2014;
Acce
pted Au
gust 20, 20
14
Resear
ch on Rock Burst Monitoring and Early Warning
Technology Based on RBF Neural Network
Yong Zhang
1
*
,
Hui Cai
2
, Yunfu Cheng
1
1
Dept of Radi
o
a
ctivit
y
,
T
a
i Shan Med
i
cal C
o
l
l
eg
e, Chin
a
2
Mine press
u
re
brunch,
Xin
g
Cun co
al min
e
, Qu fu, Shan dong, PR Ch
in
a, 2731
00,
Cha
ngch
e
n
g
R
oad 6
19,T
a
ian,
Shan do
ng, P
R
Chin
a,27
10
1
9
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: gczk
y
a
nch
e
n
@
12
6.com
A
b
st
r
a
ct
Chin
a is
o
ne
of the
most se
rious c
o
a
l
min
e
acci
de
nts in
the co
untri
es
of
the w
o
rl
d.
All of th
e
accid
ents, rock
burst is on
e o
f
them. T
he ro
ck burst
in co
a
l
and r
o
ck ma
ss, refers to the sud
den
pow
er
failur
e
, rel
eas
e
a
larg
e
nu
mb
er of c
a
tastro
phic
dyn
a
m
ic
phe
no
me
na
of
en
ergy. It ca
n b
e
destroy
the
roadw
ay roof, cause oth
e
r mi
ne dis
a
sters, casua
l
ties a
nd s
o
on. In Chi
na,
the mi
ne nu
mber w
i
th rock burst
dan
gero
u
s acc
ounte
d
for mor
e
than 20
%
of the total, Shan
don
g Qufu Xin
g
cun coal
mi
n
e
amon
g the
m
. In
order to preve
n
t
to
the hap
pen
of a
ccident, the coal
min
e
ent
erprise h
ad b
e
en insta
ll all ki
n
d
s of mon
i
torin
g
system, such a
s
SOS micro seis
mic syste
m
, Fully me
c
han
i
z
e
d
w
o
rking fa
ce resistanc
e o
f
support syst
e
m
and
so
on.
U
s
ing
sens
ors
me
asuri
n
g
an
d co
mputer
te
chno
logy, t
he
data
ha
d b
e
e
n
g
e
tting fro
m
t
h
e
und
ergro
u
n
d
1
000
meters. Accordi
ng to th
e i
n
terna
l
li
nk of
pressur
e
be
ha
vior b
e
tw
een t
he b
a
sic re
gu
l
a
rity
and v
a
ria
b
le,
RBF
neur
al n
e
t
w
o
rk had be
e
n
set up. F
r
o
m
the mod
e
l, it can forec
a
st the
risk ind
e
x of r
o
ck
burst, reve
al
th
e su
peri
n
cu
mb
ent stratu
m ro
of
move
men
t;
master
the
pr
o
c
ess of st
ate
a
nd c
h
a
nges
i
n
the
law
s
of und
er
grou
nd pr
essu
re. It is impo
rtant signi
fic
a
nce to gu
id
e
safe prod
ucti
on of coa
l
mi
ne
enterpr
ises.
Ke
y
w
ords
:
rock burst, mo
nitori
ng a
nd for
e
castin
g, RBF
neur
al
n
e
tw
ork, w
o
rking face, roof pressur
e
, micr
o
seis
m
i
c
Co
p
y
rig
h
t
©
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
With the
dev
elopme
n
t of
Chin
a, the
de
mand fo
r
co
al
mine i
s
more
and
mo
re. A
nd
coal-
related
se
curi
ty incide
nts freque
ntly occur, a
nd th
e
secu
rity situ
ation
remai
n
s g
r
im. All of va
ri
ous
types of coal
mine accid
e
n
t
s in prod
ucti
on; the roof
rock accid
ent is one of
them
. Rock burst is a
sud
den po
wer failure of
coal and rock mass, the rele
ase of large am
ounts of en
ergy
cata
strop
h
ic
dynamic
phe
nomen
a, it can be
d
e
st
roy the roa
d
way ro
of, ca
use
other
mine
disa
sters, ca
sualties a
nd si
gnifica
nt loss of life and pro
perty [1].
In recent yea
r
s, with the i
n
crea
se of m
i
ning de
pth, geolo
g
ical
co
ndition
s are
compl
e
x,
con
s
tantly improve the co
mpre
hen
sive
mecha
n
ical coal mini
ng
degree, ro
ck burst p
r
e
s
sure
appe
ars mo
re and
more p
r
omin
ent. According to t
h
e inten
s
ity, ro
ck
bu
rst i
s
u
s
ually divided i
n
to
three
cla
s
ses: mild sh
ock,
medium i
m
p
a
ct, and
st
ro
ng impa
ct of
three level
s
.
Whe
n
it occu
rs,
may indu
ce the mag
n
itud
e 3~4
earth
q
uakes, the
m
a
ximum ca
n
rea
c
h level 5
~
6. Acco
rdin
g to
the impa
ct o
r
ientation id
en
tification resu
lts, Xi
ng cun
coal
ro
ck bel
ong
s to the t
h
ird
cla
s
s, the
stron
g
impa
ct
tenden
cy.
There are m
any factors a
ffect
the birth
of rock b
u
rst, such as
nat
ural fa
ctors, tech
nical
factors, man
ageme
n
t fact
ors.
Natural factors in
cl
u
d
e
the origi
nal
rock st
re
ss,
tectonic
stre
ss,
coal
se
am c
o
ndit
i
on
s;
t
e
ch
nical f
a
ct
or
s i
n
clu
d
ing the l
o
cal
stre
ss concentratio
n
, mining
spe
e
d
,
beyond
the coal seam mining,
prevention
me
a
s
ure
s
are ina
dequ
ate; managem
ent factors
inclu
de in
ade
quate inve
st
ment, the un
reasona
ble o
p
e
ration
proce
dure
s
, respo
n
sibility he
art
is
not stro
ng, st
aff training is
not in place.
Becau
s
e
of the ha
za
rd
s of rock b
u
rst is hug
e, predictio
n of impact p
r
e
ssure, ha
s
attracte
d wid
e
attention a
ll over the
worl
d.
The
r
e
are ma
ny p
r
edi
ction met
hod
s, su
ch
as:
comp
re
hen
si
ve index
me
thod, the
dril
ling meth
od,
sei
s
mi
c m
e
thod,
sou
nd,
electroma
gne
tic
radiatio
n met
hod, test met
hod an
d so
o
n
. Many
dom
estic coal ent
erp
r
ises
in ch
ina
have
be
en
set u
p
ma
ny inde
pen
dent
monito
ring
and fo
re
ca
st
ing
system, t
hese p
r
od
uct
s
p
u
rcha
se
from
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch on
Ro
ck Bu
rst M
onitorin
g
and
Early
Wa
rnin
g Tech
nolo
g
y Based on
… (Yong Zha
ng)
7479
different
cou
n
t
ries
and
com
panie
s
. Usin
g
sen
s
o
r
me
asuring
and
co
mputer te
ch
n
o
logy, the dat
a
had be
en g
e
tting from the und
erg
r
o
und mo
re th
an 100
0 me
ters. The m
onitorin
g
system
inclu
d
ing S
O
S
micro
sei
s
mic sy
st
em,
f
u
lly
mecha
n
i
z
ed
wo
rki
n
g
f
a
ce r
e
si
st
a
n
ce
of
sup
p
o
rt
sy
st
em a
n
d
so o
n
.
B
u
t
t
hese
pro
d
u
ct
s
wit
h
dif
f
erent dete
c
t
i
on pri
n
ci
ple
s
, the dete
c
ted
information is incomplete, i
m
prec
i
s
e, in
complete
and
sometim
e
s contradi
cto
r
y. The p
a
ramet
e
rs
are
non
-lin
ea
r. Acco
rdin
g t
o
the i
n
ternal
link of
pre
ssure
beh
avior betwe
en
th
e basi
c
reg
u
la
rity
and va
riabl
e,
usin
g
RBF n
eural
net
wo
rk tech
nolo
g
y to be
the
information fu
sion
. From
the
da
ta
informatio
n, it can
reveal
the su
pe
rin
c
umbent
strat
u
m ro
of mov
e
ment;
ma
ster
the proce
ss of
state and
cha
nge
s in the la
ws of un
derground p
r
e
s
su
re. There i
s
im
portant
signifi
can
c
e to gui
d
e
safe produ
cti
on of coal mi
ne enterpri
s
e
s
.
2. Rock
Burs
t Monitoring
Principle
2.1. Micro Seismic Sy
stem
The n
e
w
g
eneration S
O
S micro
seismi
c mo
nitoring
inst
ru
ment pu
rcha
sed f
r
om
Portland. It was de
sig
n
an
d manufa
c
ture by Polis
h Mining Re
se
arch
Institute of
Mining
Inst
itute
of Seismolo
g
y
. The main purpo
se of mi
cr
o sei
s
mi
c is
predi
cting
ro
ck burst [2].
The SOS micro seismi
c m
onitorin
g
sy
stem ca
n be a
c
hieved, incl
ud
ing the rock b
u
rst o
n
mine ea
rthqu
ake
sign
als o
v
er long di
sta
n
ce
s no m
o
re than 10
km
in real time,
dynamically and
automatically monitori
ng,
getting full vibration
wa
vef
o
rm of the
ro
ck
bu
rst a
nd
mine ea
rthq
u
a
ke
sign
al. Software can a
c
cu
rately cal
c
ula
t
e the ener
gy
greate
r
than
102J of
the coal
-rock sh
o
ck
occurre
d
at a
time, inclu
d
e
energy and
three
-
di
m
e
n
s
i
onal
spa
c
e
coordi
nate
s
, to determi
ne e
a
ch
mine ea
rthqu
ake
sho
c
k type, to determ
i
ne the vibrat
i
on of the power so
urce rock mine p
r
e
s
sure
level of ri
sk
asse
ssm
ent
and fo
re
ca
st. And throu
g
h
the ap
plicati
on of the S
O
S micro seismic
monitori
ng
system, en
gine
er
can a
nalyze the mine
o
v
erbu
rde
n
fra
c
ture,
de
scrib
e
the mig
r
ati
on
of the spa
c
e rock structu
r
e
motion and
stress fiel
d evol
ution for co
al mine safety p
r
odu
ction.
The
system i
s
mai
n
ly co
m
posed of
und
er
g
r
ou
nd a
n
d
gro
und
mou
n
ted three p
a
r
ts: 16
DLM
-
20
01 d
e
t
ection p
r
ob
e
s
, floor m
oun
ted 16
cha
n
n
e
l DLM
-
SO
si
gnal a
c
q
u
isiti
on statio
n, a
n
d
AS-1 sign
al
reco
rde
r
and
so on,
th
ey compl
e
me
nt
each oth
e
r to form
a
co
mplete
syste
m
of
wor
k
.
Thro
ugh th
e relevant software, the
sy
ste
m
ca
n a
c
cura
tely calculate
the 3
D
coo
r
dinate
s
of time, ene
rgy and
space of
coal
an
d ro
ck m
a
ss ene
rgy g
r
ea
ter than
10
2
J
sh
oc
k
oc
cu
r
s
,
determi
ne th
e motion typ
e
for e
a
ch m
i
ne ea
rthq
u
a
k
e, jud
ge th
e
vibration
ge
neratin
g
sou
r
ce,
carrie
s o
n
th
e app
rai
s
al t
o
the mine
rock bu
rs
t h
a
z
ard de
gree,
can
gre
a
tly redu
ce the l
o
ss o
f
pre
s
sure
di
saster
impa
ct coal mine, g
e
tting
eno
rm
ous e
c
ono
mi
c b
enefits
an
d social
be
n
e
fits
.
Thro
ugh
the
appli
c
ation
of
SOS mi
cro
sei
s
mic mo
ni
toring
sy
stem
, it ca
n a
naly
s
is of th
e mi
ne
overbu
rd
en rock fra
c
ture informatio
n, accurate
de
scribe sp
atial strata stru
ctu
r
e
movement a
n
d
stre
ss field e
v
olution law,
serve
s
for th
e safety
in produ
ction of coal mine. Th
e re
sea
r
ch a
nd
field appli
c
ati
on re
sult
s of implementati
on will b
r
ing
a benefi
c
ial referen
c
e fo
r
rock b
u
rst an
d
other dyn
a
mi
c disaste
r
p
r
evention an
d
other
a
s
pe
ct
s, and a
c
hi
e
v
e huge e
c
o
nomic
and
social
benefits.
Figure 1. Strong Minin
g
Earthqu
ake Wave and Para
meter Chan
g
e
s
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: 747
8
– 7485
7480
In gene
ral, vibration
of ela
s
tic e
nergy relea
s
e p
r
o
c
e
ss i
s
mo
re bi
g, the vibrati
on wave
prop
agatio
n spe
ed, amplit
ude is big
g
e
r
; affect
ed by geolo
g
ical
structure and
stratum litholo
g
y
compl
e
x geo
logical environmental fa
ctors, the
r
e a
r
e vibration
wave p
r
op
ag
ation deviati
on
betwe
en in coal and rock
and the ideal
state. More
o
v
er, the vibra
t
ion freque
ncy is low, focu
sed
on the follo
wi
ng 20
Hz, vibration fre
quen
cy ra
nge
und
er spe
c
ial circum
stan
ce
s may
be slight
ly
large
r
, but in
the pro
c
e
s
s
of vibration e
nergy
atten
u
a
tion, frequ
e
n
cy ra
nge of
motion in
cre
a
sed
to a ce
rtain d
egre
e
. Most
vibration p
r
op
agation
time
betwe
en 1
~
2.5 se
con
d
, the sa
me ela
s
tic
energy rele
a
s
e in a sho
r
ter time, the e
x
tent of
damage to the co
al and rock
mass is la
rge
r
. In
the vibratio
n
energy attenu
ation
st
age, e
nergy i
s
relati
vely wea
k
, th
e vibratio
n
sp
eed
slo
w
d
o
wn,
redu
ce a
m
plit
ude, main fre
quen
cy also increa
sed.
In wave
pro
p
agation,
wav
e
velocity i
s
g
r
eate
r
, indi
cat
e
s that th
e
wave from th
e
cente
r
of
the greate
r
, the amplitud
e, the corre
s
po
nding
ki
neti
c
energy is gre
a
ter. Wh
en st
rong
ro
ck b
u
rst
energy level
more
than
10
5
J
oc
curs
, the vibr
ation wave veloc
i
ty
increases sharpl
y, the amplitu
de
of vibration is incre
a
sed to a large a
m
plit
ude, ene
rgy relea
s
e in a short pe
riod of
time.
2.2. Fully
Me
chanized Wo
rking Face
Resista
n
ce o
f
Support
The ba
sic m
ean
s of roof sup
portin
g
in
worki
ng fa
ce
s is the hydraulic
supp
ort
or sin
g
le
hydrauli
c
p
r
o
p
, emulsi
on
as the p
r
e
s
sure tra
n
sfe
r
medium, abili
ty to support
the roof ra
ck or
pillars known as the
support resi
stance, support resistance refl
ects the
roof
on intensity o
f
sup
port eq
ui
pment, pre
ssure
can be
revealed thro
ugh the medi
um of roof
pressure su
ppo
rt or
pillar
cavity.
Measurement
of hydrauli
c
pre
s
su
re me
asu
r
em
ent way have man
y
methods, t
h
is
system a
dopt
s the re
si
stan
ce st
rain me
a
s
uri
ng meth
o
d
[2].
The re
sista
n
ce is con
n
e
c
te
d with shiel
d
powere
d
su
p
port. Usi
ng st
rain ga
uge p
r
essure
sen
s
o
r
s or vi
brating
wi
re p
r
essu
re
se
nsor. The
emul
sion
of the pil
l
ars of t
he int
e
rnal
pressu
re is
delivere
d
to the se
nsor, th
e sen
s
o
r
out
put of t
he an
alog si
gnal
converted by
t
he ci
rcuit after th
e
comp
uter a
c
quisitio
n
. The
total collecti
on real
-time
pre
s
sure info
rmation in
clu
des th
ree p
a
r
ts:
the anterio
r column p
r
e
s
su
re, post
e
rio
r
column p
r
e
s
su
re and the p
r
obe be
am pressure.
The re
si
stan
ce strain
sen
s
or ha
s the ad
vantage
s of simple structu
r
e, small volu
me, high
respon
se fre
quen
cy, easy
to desig
n the stru
ctu
r
e.
It is wid
e
ly ap
plied in the
measurement
of
pre
s
sure field
.
The b
a
si
c
p
r
inci
ple
of re
sistan
ce
strai
n
se
nsor i
s
a
strain
ga
uge,
co
nsi
s
ting
of
a
bridg
e
, whe
n
the bridge a
r
m re
sista
n
ce cha
nge,
bridge lo
se
s balan
ce; the b
r
idge o
u
tput an
unbal
an
ced voltage. The o
u
tput voltage and the si
ze
of the bridge arm ch
ang
es propo
rtion
a
l
relation
shi
p
. The wea
k
out
put
voltage si
gnal
th
rou
gh
the amplifie
r
output to A/D circuit
switch
ed
by the compu
t
er acq
u
isitio
n and processing [4-5].
Figure 2. The Principl
e of Wirele
ss
Pressu
re Mo
nitoring
Step and strength of the
working fa
ce sup
port re
sista
n
ce mon
i
toring role
mainly in:
maste
r
the l
a
w of
cycle
pressure
a
nd
strata b
ehavio
r; analyze
an
d
verify the ad
aptability of the
sup
port
s
on the
roof con
d
itions; adju
s
t
sup
p
o
r
ts
re
aso
nabl
e co
ntrol
to
p state;
ro
of
a
cci
d
ent
predi
ction tha
t
may occu
r; bra
c
ket failure rate
of the
hydrauli
c
sy
stem monitori
ng, and ea
sy
to
corre
c
tly guid
e
the prod
ucti
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch on
Ro
ck Bu
rst M
onitorin
g
and
Early
Wa
rnin
g Tech
nolo
g
y Based on
… (Yong Zha
ng)
7481
Grou
nd mo
nitoring
serv
er and d
o
wn hole data
transmi
ssio
n sup
port
multiple
transmissio
n
mode
s: Ind
u
s
trial
Etherne
t bus
tra
n
sm
issi
on
mod
e
; RDS tele
ph
one li
ne
mo
de;
singl
e-m
ode
fiber tran
smi
ssi
on mod
e
. RDS commu
nicatio
n
tech
nology u
s
ing
base
ban
d type
isolatio
n tran
smissio
n
mo
de and b
a
la
nce
d
type floating comm
unication technolo
g
y; In the
telepho
ne co
mmuni
cation in
high noi
se environ
ment
with sta
b
le transmi
ssion,
cable 2
0
km, d
oes
not requi
re
sp
ecial
cabl
e laying, may consti
tute the m
o
st economi
c
monitorin
g
system.
Grou
nd
part
s
in
clud
e th
e monito
ring
se
rver,
co
mmuni
cation
interfa
c
e
a
nd
so o
n
.
Und
e
rg
ro
und
equi
pment,
incl
udin
g
fu
lly mechani
z
ed sup
port
wirel
e
ss pre
s
sure
mo
nito
ring
station; intrin
sically safe wirel
e
ss mo
nitoring
station; intrinsi
cally sa
fe monitoring
station;
intrinsi
cally safe data tran
sceiver; wi
rele
ss repe
ater; j
unctio
n
box a
nd su
ppo
rting
explosio
n-p
r
oof
power suppl
y, cable
co
mpositio
n.
Dynamic moni
toring
sy
ste
m
coal
mine
ado
pts mult
ilevel
distrib
u
ted st
ructure, in
trinsic
s
a
fety des
i
gn.
Comm
uni
cati
on ma
ster
st
ation an
d mu
lti station co
mmuni
cation
sub
s
tation th
e maste
r
-
slave
relatio
n
shi
p
, betwe
en the m
a
st
er
stati
on a
n
d
the
sub
s
ta
tion is
con
n
e
cted
by a
bus.
Comm
uni
cati
on station
fixed set add
re
ss codin
g
,
m
a
ster
follo
we
d by patrol each sub
s
tatio
n
,
sub
s
tation
re
ceive
s
the su
rvey instru
cti
ons, the
station ha
s bee
n
store
d
data frame, is can
be
transmitted to
the
comm
un
ication
statio
n. Com
m
uni
cation ma
ste
r
station
se
nt e
a
ch
data
to t
h
e
grou
nd receiving ho
st.
In rece
nt years, wi
rele
ss n
e
twork techn
o
l
ogy ha
s be
en wid
e
ly used in co
al mine. Mine
pre
s
sure m
o
nitoring
syste
m
of
wireless network to
a
band
on th
e
cable
way, to
reali
z
e
the fa
st
netwo
rk in th
e field in
stalla
tion and
re
pla
c
eme
n
t battery. The co
mm
unication
spe
ed ha
s im
pro
v
e
greatly, reduces the communi
cati
on time response.
Wi
reless
transmissi
on
reliability
get
guarantee
ef
fectively. In the effective
com
m
uni
cat
i
on rang
e, the
con
c
ent
ra
tor can
acco
rd
variou
s algo
ri
thm acqui
re t
o
optimal tran
smissio
n
ro
ute.
The
whol
e
system
ad
op
ts OL
DM
wi
rele
ss
net
wo
rk technolo
g
y
, wirele
ss
netwo
rk
comm
uni
cati
on. The
wi
rel
e
ss p
r
e
s
sure
monitori
ng
station dete
c
ts data info
rm
ation, and
th
en
transmits the
data to the
con
c
e
n
trato
r
a
c
cordi
ng t
o
ro
uting p
r
otocol. Amo
ng the
wirel
e
ss
pre
s
sure mo
nitoring
stati
on can
com
m
unicate fre
e
ly. When
a
wirel
e
ss p
r
e
s
sure
monito
ring
station failure, other nod
es ca
n self-orga
n
iz
e net
work, ch
oo
se anothe
r transmi
ssion li
nk,
improve the
reliability of network
com
m
unication.
Sensors get the pres
sure data through the
433M
Hz wi
rel
e
ss tran
smission, t
r
an
smit
ted to th
e
co
mmuni
cation
station. And
t
hen th
rou
gh t
h
e
comm
uni
cati
on su
bstatio
n
cable u
p
loa
d
to the groun
d serve
r
.
3. The Monitoring Param
e
ter
E3207
wo
rki
n
g face lo
cate
on Q
u
fu Xing
Cu
n coal
mi
ne, Shan
don
g province, el
evation
of E3207
wo
rking fa
ce
is from -1
200
m to -1
300m,
fa
ult developm
ent, dip an
gle
of co
al seam
is
0
~ 30
°, a
n
av
erag
e of 1
8
°
;
coal
se
am t
h
ickn
e
s
s is 2
.
5 ~ 7.5
m
, averag
e 7.1
5
m; to a le
ngth
of
about 84
0m, the tenden
cy
length is a
b
out 115m.
Th
e data collect
ed from SOS
micro
sei
s
mi
c
system, fully mech
ani
zed
workin
g face
resi
stan
ce
of
sup
port sy
ste
m
. Fully mechani
zed
working
face
re
sista
n
c
e
of su
ppo
rt system m
o
nitoring
pa
ra
meter i
s
p
r
e
s
s
u
re
(u
nit i
s
KN
).The
r
e
are
seven pressu
re sen
s
o
r
s
，
e
v
ery 5 min
u
tes a
data
acquisitio
n
. In
orde
r to
pro
c
ess the
data
of
conve
n
ien
c
e,
an average
of seven d
a
ily value
of sen
s
o
r
were
applied. SO
S micro
sei
s
mic
system m
onit
o
ring
pa
rame
ters
are vib
r
a
t
ion ene
rg
y (unit is
J)
and
vibration n
u
mber. Every
day,
the mining
sp
eed (unit is
m) is
re
corde
d
. As we
kn
o
w
, the ro
ck b
u
rst i
s
al
so in
conta
c
t with
the
advan
ce spe
ed.
Figure 3. The
Data of Re
si
stan
ce of Sup
port System
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: 747
8
– 7485
7482
Figure 4. The
Data of Micro Seismi
c
4. RBF
Neur
al Net
w
o
r
ks
In the p
r
edi
cti
on p
r
o
c
e
s
s of
ro
ck bu
rst, a
large
am
ount
of data
had
b
een
obtain
ed,
these
data is rand
o
m
and no
nlin
ear, an
d wa
s
clo
s
ely relate
d with the time. At the same time, the data
also h
a
s the
correl
ation be
tween e
a
ch o
t
her. T
he RB
F neural network h
a
s
stro
n
g
ly applicable
to
compl
e
x envi
r
onm
ents
an
d multi obje
c
t co
ntrol
re
quire
ment, a
nd ha
s the
cha
r
a
c
teri
stic of
approa
chin
g
any no
nlinea
r co
ntinuo
us functio
n
with
arbitrary p
r
e
c
ision,
and
th
us ve
ry suita
b
le
for ro
ck b
u
rst predi
ction
re
search [3].
As
we
kn
ow,
a radial
ba
sis functio
n
(RB
F
) n
e
u
r
al
net
work ha
s
an i
nput laye
r, a
hidde
n
layer
and
an
output laye
r.
The
neu
ron
s
in the
hidd
en
layer contain
Gau
s
sian
tra
n
sfer fun
c
tion
s
who
s
e o
u
tput
s are inve
rsel
y propo
rtional
to t
he distan
ce from the
center of the n
euro
n
.
RBF neu
ral n
e
twork i
s
the radial b
a
si
s functi
o
n
(Radi
al Basis F
u
n
c
tion) neu
ral n
e
twork,
usu
a
lly con
s
i
s
ts
of a
n
in
pu
t layer, o
ne
h
i
dden
layer a
nd o
ne
outpu
t layer. T
he
h
i
dden
layer i
s
a
grou
p of
ra
di
al ba
si
s fun
c
t
i
on, an
d e
a
ch
implicit
pa
ra
meter ve
cto
r
layer n
ode
re
lated
cente
r
and
width. Radi
al
basis fun
c
ti
on has a variety of
forms,
the general
Gau
ss fun
c
ti
on. RBF neu
ral
netwo
rk i
s
a
feed forwa
r
d network of
a goo
d
pe
rforman
c
e. It has th
e be
st
approximati
o
n
perfo
rman
ce,
with o
u
tput
weig
ht linea
r
relation
shi
p
i
n
the
stru
cture, trainin
g
m
e
thod i
s
fa
st and
easy, no lo
cal
optimum p
r
o
b
lem.The
acti
vation functio
n
of radi
al ba
sis fu
nctio
n
n
eural
network is
in betwe
en t
he input ve
ctor an
d the weight vecto
r
distan
ce
dist
as
variable. Th
e
gene
ral
expre
ssi
on fo
r the activatio
n
function [6-7]:
2
dist
-
）
（
R
e
dist
(1)
With the de
crease of dista
n
ce b
e
twe
en
the we
ig
hts a
nd the input
vector, the o
u
tput o
f
the netwo
rk i
s
increa
sin
g
.
For
RBF ne
u
r
al net
wo
rk l
e
arnin
g
alg
o
rit
h
m, there
are
three p
a
ram
e
ters nee
d to
solve:
the ce
nter
of basi
c
fu
ncti
on, varia
n
ce
and the
wei
g
hts from
hid
d
en laye
r to t
he outp
u
t layer.
Among th
em
, the radial
b
a
si
s fun
c
tion
used
is the
Gau
s
s fu
nct
i
on. Th
erefo
r
e the
a
c
tivation
function of ra
dial ba
sis fun
c
tion ne
ural n
e
twork
can b
e
expre
s
sed
as:
2
2
p
2
1
exp
)
R(x
i
p
i
c
x
c
(2)
In the formula:
c
p
x
x
is Euclide
an
norm;
c
is Gau
ss fun
c
tion center
；
is the varian
ce of Gau
s
s functio
n
.
From the
stru
cture of radial
basi
s
functio
n
neural network, it ca
n get
the network
output:
n
j
c
x
w
i
p
h
i
ij
,
2
,
1
,
2
1
exp
y
2
2
1
j
(3)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch on
Ro
ck Bu
rst M
onitorin
g
and
Early
Wa
rnin
g Tech
nolo
g
y Based on
… (Yong Zha
ng)
7483
In the formula:
T
2
1
)
,
,
,
(
p
m
p
p
p
x
x
x
x
is the P input sample
s;
P
p
,
,
2
,
1
, P is the total numbe
r of sa
mples;
i
c
repre
s
ent the network n
ode
s of the hidden laye
r ce
nters
；
ij
w
is the hidde
n layer to the output layer
weig
hts
；
h
i
,
,
2
,
1
is the numb
e
r of hidden lay
e
r nod
es.
i
y
is th
e a
c
tual
output th
e o
u
tput no
de a
nd the
input
sampl
e
s corresp
ondi
ng to
the
netwo
rk.
Let
d
is the desire
d
output value
s
of sam
p
les,
then the
variance ba
sed functio
n
can be
expre
s
sed a
s
：
2
1
m
j
i
j
j
c
y
d
p
(4)
5. RBF Model
Acco
rdi
ng ab
ove theory, A RBF mode
l had bee
n set up. This
model ha
s fo
ur input
para
m
eters.
(1) Pressu
re,
get from fully mech
ani
zed
work
in
g f
a
ce
resi
st
an
ce of
sup
port
sy
st
e
m
.
(2) Spe
ed of coal mini
ng.
(3) T
he daily vibration en
ergy.
(4) T
he daily vibration fre
q
uen
cy.
The m
odel
h
a
s
only o
ne
output p
a
ra
m
e
ter; it i
s
the
risk index
of ro
ck bu
rst. The
risk
index of rock
burst divided
into four grad
es. Each assi
gnment is 1,
2, 3, and 4.
(1)
No ri
sk, n
o
mine sh
ock or vibration e
nergy bet
wee
n
10
2
～
10
3
J, no
und
erg
r
o
u
nd
pre
s
sure beh
avior.
(2)
Wea
k
ri
sk, vibration en
ergy between
10
2
～
10
5J
, no undergroun
d
pressu
re b
e
h
a
vior.
(3) M
ediu
m
ri
sk
，～
vib
r
ation
energy betwe
en 10
2
106
J,
with the p
r
e
s
entation of th
e mine
pre
s
sure, deformatio
n
, but doe
s not affect the pro
duct
i
on
(4) High
ri
s
k
，，
vibration en
ergy betwee
n
Mine pressure ap
pea
r o
b
vious 1
0
2
～
10
8
J.
RBF mod
e
l
predi
ction
s
can
be u
s
ed in a va
riety of methods,
su
ch a
s
VC++
prog
ram
m
ing
,
matlab201
2
toolbox and
so on. No
w, we
use sp
ss17
statistics
software to re
alize
it. The softwa
r
e ha
s RBF t
oolbox. Amo
ng ten day
s d
a
ta, nine day
s as training
data, one day
as
predi
cted
dat
a. Table
1 i
s
the ori
g
in
dat
a and
RB
F
predicte
d
results. From the
p
r
edi
cted
re
sul
t
s
we can kn
ow the Exper_value and
RB
F_predi
cted
value is col
s
e to, and fits in with the actual
situation of worki
ng field, result
s indi
cat
e
that
the pro
posed mod
e
l is app
rop
r
iate
. Figure 5 is T
he
RBF mo
del
based o
n
SP
SS [8]. The
hidde
n laye
r
activation fun
c
tion i
s
Soft
max. The o
u
t
pu
t
layer a
c
tivation fun
c
tion
is identity. Figu
re 6
is
the
re
sidu
als of
RBF re
sid
ual
s; resid
ual
s e
r
ror i
s
within a
c
cept
able limits.
Table 1. The Origi
n
Data and RBF
Predi
cted Re
sults
Pr
essur
e
（
KN
）
Energ
y
（
J
）
Freque
nc
y
Speed (m)
Expert_value
RBF_ pr
edicted value
1117.00
106435.00
53.00
3.00
2.00
2.00
1092.00
43133.00
49.00
3.00
1.00
1.04
1051.00
36816.00
19.00
0.60
1.00
1.00
1055.00
73593.00
32.00
3.00
2.00
2.00
1056.00
146179.00
48.00
3.00
3.00
3.00
1076.00
80174.00
51.00
3.30
2.00
2.00
1108.00
68017.00
35.00
3.00
2.00
2.03
1118.00
63981.00
43.00
3.00
2.00
1.97
1116.00
46418.00
46.00
3.30
1.00
1.00
1123.00
95983.00
23.00
2.40
2.00
2.00
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: 747
8
– 7485
7484
Figure 5. The RBF Model
Based on SPSS
Figure 6. The
Resi
dual
s of RBF
6. Conclusio
n
The im
pa
ct o
f
early
wa
rni
ng in
dicato
rs of mi
ne
mi
cro seismi
c
pre
s
sure
is:
1)
Vibration
energy is l
e
ss than
1
× 10
4
J re
gion,
no
dang
er. T
h
e
vibration
en
ergy i
s
hig
h
e
r
than
1
× 10
4
J,
less than 1 × 10
5
J regio
n
,
dangerou
s rock burst disaster
, is a
weak
sho
c
k ha
zard area. And
vibration e
n
e
r
gy is la
rge, d
ange
rou
s
ro
ck bu
rst disast
er i
s
bigg
er;
2) Th
e st
ron
g
strata
beh
avior
occurre
d
bef
ore, mine
earthqua
ke
s and
seismic e
n
e
r
gy increa
se
s rapidly, maint
a
ined at a hi
gh
level, until th
e occu
rren
ce
of strong
m
i
ning la
rge
p
r
essu
re, mi
n
e
ea
rthqu
ake
s
an
d
sei
s
mic
energy re
du
ces the
mi
cro
sei
s
mi
c
sign
al; 3) F
r
e
que
ncy firstly increa
se
s g
r
ad
u
a
lly, and the
n
bega
n
to de
cl
ine sha
r
ply, whe
n
mi
cro seismi
c sign
al freque
ncy
i
n
crea
sed
ag
ain,
su
gge
sting
that
there may bu
rst.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch on
Ro
ck Bu
rst M
onitorin
g
and
Early
Wa
rnin
g Tech
nolo
g
y Based on
… (Yong Zha
ng)
7485
Usi
ng
RBF n
eural
network technol
ogy, differ
ent info
rmation
ca
n
be fusi
on. From the
data informati
on, it can rev
eal the sup
e
ri
ncum
bent st
ratum roof mo
vement; mast
er the pro
c
e
s
s
of state and cha
nge
s in the la
ws of underground p
r
essu
re. The
r
e is importa
n
t
significan
c
e
to
guide
safe produ
ction of coal mine ente
r
pri
s
e
s
.
Ackn
o
w
l
e
dg
ements
This work is suppo
rted by the Natu
ral Sc
ience Foun
da
tion of Shandong Provin
ce,
China
(Nu
m
be
r: Z
R
2011EL
019
).
The State Ad
ministrati
on o
f
safety supe
rvision
an
d
manag
eme
n
t of
proje
c
t 201
3-2014.
Referen
ces
[1]
Z
hang Yo
ng, Cai hu
i, Yang
Yong-ji
e. Stud
y
o
n
mod
e
l
of rockburst based o
n
infor
m
ation fusi
o
n
technology
.
Jo
urna
l of safety scienc
e an
d techno
logy.
2
013
; 9(3): 40-45.
[2]
Yong
Z
h
a
ng, Z
hao
jun
L
i
u, Yu
nfu C
hen
g
. Stu
d
y o
n
Mu
lti-se
nsor Infor
m
atio
n F
u
si
on T
e
c
h
nol
ogy
in
th
e
Dyna
mic
Mo
ni
toring
of C
o
a
l
Mine
R
oof
. Internati
o
n
a
l c
o
nferenc
e W
I
SM 20
12. C
h
e
n
gdu.
Lectur
e
Notes in com
p
uter scienc
e 75
29:P62
9
-6
36.
[3]
Er MJ, W
u
SQ, Lu JW
. T
oh
HL.F
ace r
e
cog
n
itio
n
w
i
t
h
ra
di
cal
basis
funct
i
on
(RBF
) n
e
u
r
al N
e
t
w
ork
.
IEEE Trans Neural Netw.
2002; 13(5): 69
7-7
10.
[4]
JIA Rui-sh
eng,
SUN H
ong-m
e
i, YAN
Xi
an
g-
hon
g. Mod
e
l fo
r safet
y
eval
ua
tion of c
oal m
i
ne ro
of bas
e
d
on evi
d
e
n
ce fu
sion the
o
r
y
.
Jo
urna
l Of China
Coal S
o
ciety,
2
010; 35(
9): 149
6-15
00.
[5]
Z
hang
Yon
g
. S
t
ud
y o
n
i
n
tegra
t
ed tech
nol
og
y
base
d
o
n
ro
of
d
y
nam
ic mo
nit
o
rin
g
. Qing
Dao
:
Shan
do
n
g
univ
e
rsit
y
of scienc
e an
d tech
nol
og
y. 20
09.
[6]
Han HG, Che
n
QL, Qiao JF
. An
efficient self-orga
n
izi
n
g RBF
neura
l
net
w
o
rk for w
a
t
e
r qu
alit
y
pred
iction.
Ne
u
r
al Netw
orks.
2011; 24(
7): 717
-725.
[7]
Buzzi C, Cr
ip
po L, Sc
ian
d
r
one M.
Conv
e
r
gent d
e
com
p
ositio
n tech
niq
ues for tra
i
ni
n
g
RBF
n
eura
l
net
w
o
rks.
Ne
ur
al Co
mputatio
n
. 2001; 13(
8): 1891-
192
0.
[8]
HUANG Hu
a. Simulati
on T
e
st of Principa
l C
o
mpo
nent A
nal
ysis
RBF
Neur
al Net
w
o
r
k Mo
del B
a
sed
o
n
SPSS.
Journal
of Xinji
ang A
g
ri
cultura
l
Univ
er
sity.
2010; 33(
6): 535-5
38.
Evaluation Warning : The document was created with Spire.PDF for Python.