Application of Rough Fuzzy Clustering Means To Earthquake Fire
Application of Rough Fuzzy Clustering Means To Earthquake Fire
Application of Rough Fuzzy Clustering Means To Earthquake Fire
LI Bo
1
, HOU Zunze
2
& ZHANG Dongdong
1
(1 Department of Fire Engineering, Institute of Chinese Peoples Armed Force Police Academy, Langfang 065000, Hebei, China;
2 Institute of Chinese Peoples Armed Force Police Academy, Langfang 065000, Hebei, China)
Abstract: On the base of the historical data, the characteristics of the earthquakes and the losses in the post-earthquakes fire are classified by using the rough fuzzy
clustering means, and then the method is given to forecast and appraise the losses in the post-earthquakes fire, and obtain a more precise result of forecast. The
algorithm presents a new objective function which incorporates the concept of the upper approximation and the lower approximation in rough sets, produces better
results than Fuzzy C-mean algorithm at time complexity. The better effect can be testified by many experiments.
Keywords: rough fuzzy C means; fire after earthquake; disaster prediction
1 Introduction
The earthquakes lead to a large number of disasters, among the secondary disasters, fire is usually the most dangerous. It may
cause an uncontrolled fire, because of the fire fighter encountering breakage. In order to decrease the loss of the fire after earthquakes, it
is very important to do some analysis of the fire after earthquakes. Some Japanese scholars do creative work on predicting and
appraising the fire after earthquakes. They point out that the model of the fire after earthquakes is related to many factors such as the
season, the time, the number of houses, the using rate of gas appliances. They classify the fire after earthquakes and analyze the reasons
of the fire after earthquakes
[1]
. It gives a method of predicting loss of earthquake-fires with the fuzzy C means .Base on above analysis,
this present study used the rough fuzzy C means to predict and assess the loss caused by fire in earthquake around the urban areas
[2]
.
This new method is easy to calculate and need less date. What more, it has self-adaptation capacity.
This paper puts forward the main idea in rough fuzzy C means algorithm including: (1) Applying rough fuzzy C-means to cluster
the earthquake fire data, different clustering centers means different degree of loss (2) Find the centers of all the clusters, computing the
corresponding loss; (3) According to the currently earthquake fire data to compute the distance to clustering center, and present the
fuzzy sets of corresponding loss based on distance.
2 Rough fuzzy C means
2.1 Basic knowledge
Suppose array
1 2
{ , , }
n
X x x x L is a finite date sets in mold space
R
p
,
1 2
( , , )
p
k k k kp
x x x x R L is called eigenvector or pattern
vector(or simply object),
kj
x is the jth character of pattern vector
k
x .
Definition 1.Let
1 2
{ , , }
n
X x x x L is the clustering set, for the ith cluster
i
w , he core is
i
v , define
{ | , 0}
i j j i i i
R w x x v A A
>
, we
have
(1)
{1 , 2 , , } , k n k j L
,
,
j i j k
x R w x R w
if
j i
x R w
,.
j i
x R w
j k
x R w
(2) At least
, if
j i
x R w
.
i
A is called the upper approximation, which describes the boundary of all the objects that may belong to the ith group. If some certain
objects do not belong to the scope the upper approximation described, they will certainly belong to the negative scope of the group.
Definition 2. The object function of RFCM is:
2
1 1
( , )
j i
N c
m
m ij ij
x R w
j i
J U V u d
Where
2
2
ij j i
d x v ,
1 2
{ , , }
c
V v v v L ,
i
v is called the clustering center of the ith cluster, weight (1, ) m .
The constraint condition is:
193
1
1
(1) : [0,1]
(2) : 0
(3) : 1
j i
ij
n
ij
j
c
ij
x Rw
i
u
u N
u
< <
'
_
,
(1)
The core equation is:
1
1
N
m
ij j
j
N i
m
ij
j
u x
v
u
(2)
We can easily get from RFCM:
1
1
j i i j
x R w u
2
FCM RFCM
ij ij
u u
We just consider three factors in this article: the number of the broken houses, the number of the semi-broken houses, and the
using rate of gas appliances .In a given records i (i=1, 2, , n) of earthquake historical dates, pi can be regard as the number of the
broken houses, qi can be regard as the number of the semi-broken houses,
i
r denotes the using rate of gas appliances,
i
s is the
quantities of the houses which have been burnedSuppose there are n historical dates and } , , , {
2 1 n
x x x X is the specimen room,
let ) , , (
i i i i
r q p x is the ith recorder.
i i i
r q p , , are the characters of the earthquake.
2.2 Algorithm steps and its analysis
The calculating step is as following:
Step 1. Calculate the center matrix
1) Confirm the parameters, (2 ) c c N ,
m
and the initialize matrix ( )
ij
C N
U u
, the limit of the upper approximation
i
A , s=0 (s
denotes the iterative time).
2) Compute the core
( )
{ }
s
i
v , according to (2).
3) If
i j
w R x
, let
0
i j
u
; else update
( ) s
ij
u according to (1).
194
4) If
( ) ( 1) s s
U U
+
< , stop; else, 1 s s + , go to 2).
Step 2. Clustering according to the distance
For
j
x X , if
{ }
1
min ij kj
k C
d d
, let
j
x belong to the
i
.
Step 3. Suppose
i
concludes } , , , {
2 1 r
i i i
x x x , and their corresponding loss is
} , , , {
2 1
r
i
i i
s s s
. Computing the loss of the
clustering center according to:
r
j
i
w
r
j
i
s
i
w
i
u
j
j j
1
1
(3)
Where
)
|| ||
exp(
d
v x
w
i j
i
j
denotes the weight, and
||} {|| max
, 1
j i
c j i
v v d
.
Which denotes the near to the core, the greater the weight is.
Step 4. Let } , , , {
2 1 c
u u u U is consisted of object, named universe, through which the fuzzy set can estimate the loss of fire.
Suppose ) , , (
1 1 1 1 + + + +
n n n n
r q p x is the character of someone earthquake, we want to estimate the loss of this earthquake fire
1 + n
s .
We can estimate
1 + n
s according to the following fuzzy sets.
c
u
c
a
u
a
u
a
n
s + + +
+
2
2
1
1
1
~
(4)
Where
) , , 2 , 1 ( ,
1
)
||)
1
|| 1 (
1
(
||)
1
|| 1 (
1
c i
c
k
k
v
n
x
i
v
n
x
i
a
+
+
+
+
Notice: (1) The former two steps progress clustering based on RFCM, the after two steps put forward the method to estimate the
earthquake fire. (2) In step 3, we get
c
u u u , , ,
2 1
, which is the ranking partition. Different clustering means different ranking. The
number of clustering determines the estimation precision. (3) From step 4, we can easily get that the near the object to the clustering
center, the greater the degree of membership as loss. (4) As for some historical data of earthquake fire, if there are some new data, the
clustering center will change, which indicate this algorithm has self-adaptive.
3 Data Analysis
We have gotten some data of historical earthquakes which are shown in Table 1
[1]
.
Table 1 Losing cases in historical earthquakes
No. Time Place Broken Semi-broken Burned houses
1 1914.03.15-04:59 JapanJiTian 640 575 3
2 1923.09.01-11:58 GuanDong 128266 126233 447128
3 1925.05.23-11:09 DanMa 309 271 148
4 1927.03.07-18:28 DanHou 1861 1666 416
5 1930.11.26-04:03 YiDou 2165 5516 75
6 1936.05.01-14:58 NanLu 479 858 9
7 1943.09.10-17:57 NiaoQu 7485 6158 251
195
8 1946.12.21-04:19 NanHaiDao 11591 23487 2598
9 1948.06.28-16:13 FuJing 36184 11816 3851
10 1952.03.04-10:23 ShiShengJinHai 1614 5449 20
11 1964.06.16-13:01 XinKe 1960 6640 290
12 1968.05.16-09:49 QingSen 673 3004 13
13 1970.10.16-14:26 QiuTian 0 20 1
14*