You are on page 1of 8

Expert Systems with Applications 39 (2012) 62136220

Contents lists available at SciVerse ScienceDirect

Expert Systems with Applications


journal homepage: www.elsevier.com/locate/eswa

Fuzzy risk analysis of ood disasters based on diffused-interior-outer-set model


Qiang Zou a, Jianzhong Zhou a,, Chao Zhou b,, Jun Guo a, Weiping Deng a, Mengqi Yang a, Li Liao a
a
College of Hydropower & Information Engineering, Huazhong University of Science and Technology, Wuhan, 430074, PR China
b
Department of Control Science and Engineering, Huazhong University of Science and Technology, Wuhan 430074, PR China

a r t i c l e i n f o a b s t r a c t

Keywords: Floods are indeed one of the most serious natural hazards for human societies, especially in China. In this
Flood disaster paper, we rstly introduce the interior-outer-set model (IOSM) based on information diffusion theory in
Fuzzy risk analysis detail. Then taking consideration its deciencies, we represent the diffused-interior-outer-set model
Information diffusion (DIOSM) to obtain the possibilityprobability distribution (PPD). Based on the PPD, we analyze and cal-
Diffused interior-outer-set model
culate the risk results. To illustrate the procedure of the proposed method, we apply DIOSM to describe
Possibilityprobability distribution
Flood risk management
the ood disaster risk quantitatively in China by using statistical data respectively, such as the time of
oods, the number of the deaths as well as the economic losses each year from 1990 to 2009. The out-
comes of this research offer new insights and moreover new possibility to carry out an efcient way
for various future ood disaster prevention and mitigation.
 2011 Elsevier Ltd. All rights reserved.

1. Introduction more and more attention within country, for it can provide infor-
mation on previous, current and future ood risks (Wang & Plate,
Flooding is not just a regional problem but a key facet of global 2002; Wei, Jin, & Yang, 2002; Zhang, Zhou, Xu, & Watanabe, 2002).
water management and the risk it poses is widely ranging. Flood Uncertainty always exists, and therefore, risk is inevitable
hazard, risk, and disasters are the products of an interaction (Karimi & Hllermeier, 2007). Besides, future ood variations are
between environmental and social processes (Parker, 2000). an issue of great uncertainty, and the risk analysis of oods is abso-
Worldwide statistics indicate that continuously increasing ood lutely necessary (Feng, Hong, & Wan, 2010). There are many risk
damages and losses of human lives remain at high levels unaccept- analysis models at present in use that attempt to evaluate and esti-
ably. It is estimated that losses caused by ood account for 40% of mate risk. In order to provide valuable support for decision making,
all total losses attributed to all disasters. Obviously, ood disasters most existing risk analysis models based on quantitative tech-
are one of the most frequent and devastating disasters over the niques rely more on statistical calculations than judgment and give
world (International Federation of Red Cross & Red Crescent Soci- highly subjective results, such as Monte Carlo, Failure Mode and Ef-
eties, 1998). However, in societal terms, the main emphasis is fects Analysis, Sensitivity Analysis (White, 1995), Simulation and
the risk to people and property. It is neither practically nor eco- Annual Loss Expectancy (Ngai & Wat, 2005), Fault-tree analysis
nomically feasible to eliminate all ood risk. Therefore, the most (Cheng, Lin, Hsu, & Shu, 2009), Annual Loss Expectancy (Rainer,
suitable approach for dealing with ooding must be to manage Snyder, & Carr, 1991). However, these traditional approaches have
the risk best (Fleming, 2002). their disadvantages when the available data are insufcient to per-
The risk of ooding is a major concern in many areas nearly all mit estimating reliably the probabilities of release of risk agents or
around the world and especially in China. Up to two-thirds of the other characteristics of concern. In many cases, risks are rather fuz-
land in China are at different types and levels of risk from river zy for our perception because of the shortage of knowledge or
and coastal ooding, which results from the combined effects of information about the systems that determine the adverse inci-
natural, social and economic factors. Floods cause huge economic dents (Huang & Ruan, 2008). If pure probabilistic methods are em-
damage and destruction, and the economic losses are about ployed to analyze the risk, unreliable results would be obtained, so
2228.5 billion, which is about 33.343.2% of the natural disaster fuzzy methods to analyze the risk of natural hazards are proposed
losses every year. What is more, oods occur quite frequently, for in succession (Huang, 1996).
there being a larger disastrous ood every 23 years in China since Fuzzy mathematics has been widely applied in natural disaster
1949. In addition, the span of oods every year is very long. So it is studies to deal with uncertainty problems. After the fuzzy set theory
commonly believed that ood risk analysis is suitable and receiving was established by Zadeh (1978), many scholars have used fuzzy
mathematical methods to study risk assessment. Misra and Weber
Corresponding author. Tel.: + 86 0278 7543127. (1990) used the fuzzy set theory for level-1 studies in probabilistic
E-mail address: jz.zhou@hust.edu.cn (J. Zhou). safety assessment and risk management. Warmerdama and Jacobsa

0957-4174/$ - see front matter  2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.eswa.2011.12.008
6214 Q. Zou et al. / Expert Systems with Applications 39 (2012) 62136220

(1994) presented a mathematical model for optimally siting and only lie within a certain interval Ij. For the sake of simplicity, we
routing hazardous waste operations. Fisher (2006) adopted fuzzy always use uj to represent Ij. When a sample point is subjected to
decision-making frameworks to improve environmental quality in random disturbance, it may depart from the interval Ij, while a
view of the uncertainty. Fu (2008) used a fuzzy optimization meth- point outside may also enter the interval. The possibility of sample

od to solve multi-criteria decision making problems under fuzzy point xi leaving or joining the interval Ij is expressed as q
ij and qij ,
environments. Bardossy, Bogardi, and Duckstein (1991) proposed respectively. For the purpose of computation, the interior set and
the fuzzy-set risk analysis technique which is also considered as outer set of Ij is dened by Huang (2004) as follows:
an extension of interval analysis, to combine and rank fuzzy risk
estimates for risk management. Lee and Chen (2008) presented a (1) Xinj = X \ Ij is called an interior set of interval Ij. The ele-
new method for fuzzy risk analysis based on fuzzy numbers with ments of Xinj are called the interior points of Ij
different shapes and different deviations. All these studies above (2) Xoutj = XnXinj is called an outer set of interval Ij. The ele-
were formalized by using the fuzzy set theory, or combined with ments of Xoutj are called the outer points of Ij
other theories to enhance the research for fuzzy risk analysis. But
a less common approach is to employ a risk calculation model to "xi 2 X, if "xi 2 Xinj, we say that it loses information, by gain at
deal with the inherent risk of the fuzzy treatment exhaustively 1  qij, to other interval, and we use q ij 1  qij to represent the
(Reyna & Brainerd, 2008). loss; if "xi 2 Xoutj, we say that it gives information, by gain at qij
And Huang (1997) established information diffusion theory to Ij, we use q ij qij to represent the added information. So for Ij,
which employed fuzzy set theory to complement the probability with information gain qij, these possibilities can be formulated as
theory with an additional dimensional of uncertainty, can charac- follows Eqs. (2) and (3):
terize the fuzziness of the probability by a possibility distribution 
1  qij ; if xi 2 X inj
and enhanced its effectiveness in small sample optimization with qij 2
inherent imprecision and a scarcity of statistical data (Feng et al., 0; if xi 2 X outj
2010; Karimi & Hllermeier, 2007). The interior-outer-set model 
(IOSM) rst established by Huang based on information diffusion qij ; if xi 2 X outj
qij 3
theory, is capable of calculating the possibilityprobability distri- 0; if xi 2 X inj
bution (PPD) values in the occurrence of oods to solve multi-val-
qij is called the leaving possibility, and qij the joining possibility.
ued risks. And the possibilityprobability risk calculated using the
The leaving possibility of an outer point is dened as 0 (for it has
IOSM is referred to as fuzzy risk. IOSM has been using to assess the
gone). The joining possibility of an interior point is dened as 0
fuzzy risk of natural disaster, and the results are demonstrated bet-
(for it has been in the interval).
ter than the conventional probability methods in risk management
Let Q  
j fqij g be the list of membership degrees to the informa-
(Huang, 2004). After all, IOSM also has three deciencies (Huang,
tion diffusion distribution from sample points within the observa-
Zong, & Chen, 2007), so we improve it to be a new model called dif-
tion interval and Q
j fqij g be the list of membership degrees to
fused-interior-outer-set model (DIOSM), and we employ DIOSM to
the information diffusion distribution, from sample points outside
calculate the fuzzy risk of ood disasters in China. The results for
the observation interval. If jQ  j j nj , we can use Eq. (4) to calculate
DIOSM reect the ood risk in China effectively, which would pro-
a possibilityprobability distribution (PPD), which is called the
vide the government and decision-makers with a more suitable
interior-outer-set model (IOSM).
and invaluable guide and overview on ooding. Therefore, it is rec-
8
ommended to study ood disasters using DIOSM in China.
> 1stsmallest element of Q j ; if p 0
In this paper, DIOSM focusing on providing information to help >
>
>
>
make better decisions in an uncertain world, is applied to calculate
>
> 2ndsmallest element of Q j ; if p 1n
>
>
>
>
the fuzzy risk of ood disaster. And the paper is organized as >
> 
>
>
follows. In Section 2, we rstly introduce IOSM briey, and then >
>  nj 1

present the measurements to improve it, thus propose DIOSM. > Lastlargest element of Q j ; if p n
>
>
< n
There is a case study in Section 3 to show the process of DIOSM. pIj p 1; if p nj 4
>
>
Finally, we discuss and make in Section 4. >
> n 1
>
> 1stlargest element of Q j ; if p jn
>
>
>
> n 2
>
> 2ndlargest element of Q j ; if p jn
2. Diffused interior-outer-set model >
>
>
>
>
> 
>
>
2.1. Interior-outer-set model :
Last smallest element of Q j ; if p 1

First we introduce the interior-outer-set model (Huang & For a size n, the universe of discourse of probability is
Moraga, 2002). Let X = {xiji = 1, 2, . . . , n} be a given sample, xi 2 R  
1 n1 n
(set of real numbers), and U fuj jj 1; 2; . . . ; mg  R be a discrete P fp0 ; p1 ; . . . ; pn g 0; ; . . . ; ; 5
n n n
universe of X, with a given step length D, D  uj  uj1,
j = 2, 3, . . . , m. From the point of view of the information distribu- And for the sample X, using Eq. (6), we can obtain a PPD,
tion, a sample point xi can allocate its information with a value pIj p; j 1; 2; . . . ; m; p 2 P, that is a fuzzy relation on the Cartesian
of qij to point uj, called one dimensional linear information-distri- product space {I1, I2, . . . , Im}  {p0, p1, . . . , pn}, written as
bution shown in Eq. (1):

1  jxi  uj j=D; if jxi  uj j 6 D
qij xi 2 X; uj 2 U 1
0; else

where xi is the observation value, uj is a controlled point, and D is


the step length of controlled point. qij is called an information gain.
Let Ij buj  D=2; uj D=2 be the observation interval associ-
ated to uj, intervals are selected so that all of the sample points xi 6
Q. Zou et al. / Expert Systems with Applications 39 (2012) 62136220 6215

The IOSM has three disadvantages (Huang et al., 2007): (1) Se- 2.4. Calculation with a PPD
lect the domain and step length of its universe arbitrarily just
depending on personal instinct and experience, and the informa- In a PPD, the number of intervals is m, while the number of
tion taken by a sample point is merely distributed to its two adja- probabilities for each interval is n + 1. We may consider that for
cent intervals. Nevertheless, the more intervals are set, the more a interval Ij, there are n + 1 probabilities pIj pi i 0; 1; . . . ; n for
concentrated the information distribution is. (2) The membership it, vice versa. So in order to obtain as much information as possible
of this model is less than or equal to 0.5 if it is not up to 1. This for risk management, we calculate the weights and expectation of
means the gradient descent of the membership function which probabilities in the interval Ij using Eq. (13) as follows:
shows a fuzzy probability is sharply changed. (3) Many values of X
n
the PPD take 0, making the information too concentrative. Euj xi pi 13
In order to improve the IOSM, we employ the asymptotical opti- i0
mal formula and the distribution of the original data to obtain a
where the weight xi (i = 0, 1, . . . , n) is calculated as follows (Liu &
more reasonable step length of universe. This can be considered as
Liu, 2002; Shen & Zhao, 2010):
the rst advantage. And the diffusion function instead of the linear

distribution function is applied to diffuse the information carried by 1
each sample point, which can solve the last two deciencies. Based x0 l0 max lr  max lr 14
2 06r6n 0<r6n
on the above improvements shown in Sections 2.2 and 2.3, we
1
obtain the Diffused-IOSM (DIOSM). Huang (2004) demonstrated xi max lr  max lr max lr  max lr ; 1 6 i 6 n  1 15
2 06r6i 06r<i i6r6n i<r6n
that DIOSM is more reasonable and accurate than IOSM.
1
xn max lr  max lr ln 16
2.2. Selecting a step length of the universe 2 06r6n 06r<n

where
Pn
lr pIj pr ; r 0; 1; . . . ; n. Obviously, xi P 0 and
First, we apply Eqs. (7) and (8) to calculate the number of inter-
i1 xi maxi6r6n lr 1.
vals m with the sample size n:
According to Eq. (13), we use the following Eq. (17) to calculate
m0 1:87  n  10:4 7 the average probability of that event x 2 Ij occurring.
,
m m0 k  dX 8 X
m
puj Euj Euj ; j 1; 2; . . . ; m: 17
where k and d(X) are used to adjust m0 to be a integer m, and When j1
X is density, d(X) = 1, otherwise d(X) = 1.
Then, rearranging sample X = {xiji = 1, 2, . . . , n} in ascending or- Thus we obtain the probability distribution p(uj), respectively.
der, we obtain X  xi jx1 6 x2 6    6 xn ; xi 2 X . Then, we dene the exceeding probability estimation
Finally, in order to make full use of the information, we focus on P = {P(uj)jj = 1,2, . . . , m} as the risk estimation value:
study interval I x1  D2 ; xn D2 , where D is the step calculated by X
n

Eq. (9):
Puj Pu P uj puk 18
 kj
D xn  x1 =m  1 
9
Finally according to Huang (2004), we employ the following
Eqs. (19)(21) step by step to obtain the total fuzzy expected value
2.3. Replacing the distribution by the diffusion E for the PPD by the so-called center-of-gravity method as follows:
,
X
n X
n
Let x0 be a sample point, we apply the normal diffusion function cx 2 Ij pk  pIj pk pIj pk ; j 1; 2; . . . ; m: 19
(Huang et al., 2007) shown in Eq. (10) to improve IOSM, so that the k0 k0
information taken by any sample point can be diffused on the ,
whole universe in a form of a diffusion function. Xm
! px 2 Ij cx 2 Ij cx 2 Ij ; j 1; 2; . . . ; m: 20
1 v  x0 2 j1
lx0 ; v p exp  10
2pr 2r2
 X
m
where r d=U a=2 ; d xn  x1 =2 is called the radius of X, Ua/2 is E uj  px 2 Ij 21
the upper a/2 critical fractile of normal distribution function. j1

And the information that interval Ij obtained from sample point


xi is 2.5. Diffused interior-outer-set model (DIOSM)
Z
qij lxi ; v dv 11
v 2Ij In this paper, the process of DIOSM can be described as follows:

As we know, the domain V is not often the real set, so the infor- Step 1: Calculate the number m of intervals with Eqs. (7) and (8),
mation taken by sample points can be lost inevitably when dif- then obtain the step D with Eq. (9)
fused on the whole universe. And the diffusion function satises Step 2: Using Eqs. (10) and (11) instead of Eq. (1) to calculate the
information approximate conservation law, i.e., as any sample distribution information qij.
point at the midpoint of a diffusion interval, Step 3: Calculate the leaving information q
! ij and qij with Eqs. (2)
Z and (3) respectively, then obtain the Q  and Q
x0 d
1 v  x0 2 j j
p exp  dv 1  a  1 12 respectively
x0 d 2pr 2r2
Step 4: Using Eqs. (4)(6) to get the obtain the fuzzy risk calcu-
Here we impose the a = 0.01, so Ua/2 = 2.58, and loss informa- lated by the PPD, which is the complexity and main step
tion is generally restricted to be less than 1%, ensuring that its in the model.
information could be approximately conserved over the whole Step 5: For the PPD, calculate the weights and expected value of
study area (Li, 1994). probabilities in every interval uj with Eqs. (13)(16).
6216 Q. Zou et al. / Expert Systems with Applications 39 (2012) 62136220

Step 6: Using Eqs. (17) and (18) to calculate the probability p(uj) Table 3
and risk estimation value P(uj) respectively, and nally cal- Top 10 natural disasters in China for the period 19002010 sorted by economic
damage costs.
culate the total fuzzy expected value E with Eqs. (19)(21).
Disaster Date Damage (000 US$)
Earthquake (seismic activity) 12-May-2008 85,000,000
3. Case study Flood 1-Jul-1998 30,000,000
Flood May-2010 26,085,410
Extreme temperature 10-Jan-2008 21,100,000
According to The International Disaster Database (http:// Drought Jan-1994 13,755,200
www.emdat.be/), we obtain the statistical data on the occurrence Flood 30-Jun-1996 12,600,000
and impact of the top 10 natural disasters in Peoples Republic of Flood 23-Jun-1999 8,100,000
Flood 23-Jun-2003 7,890,000
China since 1900, such as drought, earthquake (seismic activity),
Flood 1-Jun-1991 7,500,000
epidemic, ood, storm and so on, as shown in Tables 13. It is Flood 15-May-1995 6,720,000
unquestionable that ood is the main natural disaster, which
Annotation: data source is the international disaster database (http://www.em-
not only is a huge threat to lives and property, but also puts a
dat.be/).
constant risk to human beings, for the number of people killed,
injured or affected and disaster-related economic damage esti-
mates are huge (Mikhailov, Morozov, & Cheroy, 2008). And with
Table 4
the booming of the economy, consequences resulted by oods be- The time of oods, the number of deaths, and the economic losses each year from
come larger and lager. It is easily seen that the majority of the 1990 to 2009 in China.
data on ood disaster from the tables is in the year after 1990.
Year Number of Number of Economic losses (million
It is so important that putting more emphasis on ood risk oods deaths US$)
analysis.
1990 3 495 803
Based on DIOSM proposed above, we analyze the fuzzy risk of 1991 3 1861 8044.12
ood disasters and calculate the ood risks quantitatively in China 1992 6 540 509.1
by using three series of statistical data respectively, such as the 1993 4 1231 6088
time of oods, the number of the deaths as well as the economic 1994 6 1564 7539.1
1995 3 1618 6720
losses each year from 1990 to 2009, shown briey in Table 4.
1996 4 4091 18914.5
And we consider that the results could give the public as much 1997 6 1090 2439.6
as quantitative information to make contributions to the future 1998 5 4250 31743.52
disaster prevention and mitigation. 1999 6 1185 9254.4
2000 9 422 315.366
2001 8 402 711
Table 1 2002 10 1246 4728.8
Top 10 natural disasters in China for the period 19002010 sorted by numbers of total 2003 6 662 15329.64
affected people. 2004 9 450 1985.5
2005 11 578 4660.86
Disaster Date No. Total affected 2006 20 430 1687.5
2007 12 1030 5023.155
Flood 1-Jul-1998 238,973,000
2008 7 352 2392
Flood 1-Jun-1991 210,232,227
2009 7 209 1373
Flood 30-Jun-1996 154,634,000
Flood 23-Jun-2003 150,146,000 Annotation: data source is the international disaster database (http://www.em-
Flood May-2010 134,000,000 dat.be/).
Flood 15-May-1995 114,470,249
Flood 15-Jun-2007 105,004,000
Flood 23-Jun-1999 101,024,000
Flood 14-Jul-1989 100,010,000 3.1. Fuzzy risk analysis using the time of oods based on DIOSM
Storm 14-Mar-2002 100,000,000

Annotation: data source is the international disaster database (http://www.em- First, we employ DIOSM to analyze the time of oods from 1990
dat.be/). to 2009 each year in detail.

Step 1: From Table 4, we get the size of sample points is 20,


X1 = {x1, x2, . . . , x20} = {3, 3, 6, 4, 6, 3, 4, 6, 5, 6, 9, 8, 10, 6, 9, 11,
20, 12, 7, 7}, and use Eqs. (7) and (8) to obtain the number
m of intervals:
Table 2
Top 10 natural disasters in China for the period 19002010 sorted by numbers of
killed. m0 1:87  20  10:4 6:0721

Disaster Date No. killed So we obtain m = 6


Flood Jul-1931 3,700,000 And x1 min16i620 xi 3; xn max16i620 xi 20, then obtain the
Drought 1928 3,000,000 step D with Eq. (9):
Flood Jul-1959 2,000,000
Epidemic 1909 1,500,000

D xn  x1 =m  1 17=5 3:4
Drought 1920 500,000
Flood Jul-1939 500,000
Earthquake (seismic activity) 27-Jul-1976 242,000 For the number of times is a integer, here we obtain D = 4. Hence,
Earthquake (seismic activity) 22-May-1927 200,000 we obtain a discrete universe of discourse of X:
Earthquake (seismic activity) 16-Dec-1920 180,000
Flood 1935 142,000 U 1 fu1 ; u2 ; . . . ; u6 g f2; 6; 10; 14; 18; 22g
Annotation: data source is the international disaster database (http://www.em-
dat.be/). So the corresponding intervals are:
Q. Zou et al. / Expert Systems with Applications 39 (2012) 62136220 6217

Table 5 I1 0; 4; I2 4; 8; I3 8; 12; I4 12; 16;


Distributed information of each sample point.
I5 16; 20; I6 20; 24
X1 u1 = 2 u2 = 6 u3 = 10 u4 = 14 u5 = 18 u6 = 22
Step 2: Using Eqs. (10) and (11) instead of Eq. (1) to calculate the
x1 0.3257 0.2737 0.1147 0.0239 0.0025 0.0001
x2 0.3257 0.2737 0.1147 0.0239 0.0025 0.0001
distribution information qij, and the results are shown in
x3 0.2351 0.3328 0.2351 0.0828 0.0145 0.0013 Table 5.

x4 0.3051 0.3051 0.1522 0.0378 0.0046 0.0003 Step 3: Calculate the leaving information qij and qij with Eqs. (2)
x5 0.2351 0.3328 0.2351 0.0828 0.0145 0.0013 and (3) respectively, then obtain the Q j and Q

j respec-
x6 0.3257 0.2737 0.1147 0.0239 0.0025 0.0001
x7 0.3051 0.3051 0.1522 0.0378 0.0046 0.0003
x8 0.2351 0.3328 0.2351 0.0828 0.0145 0.0013
x9 0.2737 0.3257 0.1933 0.0572 0.0084 0.0006 Table 8
x10 0.2351 0.3328 0.2351 0.0828 0.0145 0.0013 Fuzzy risk of the time of oods represented by the PPD.
x11 0.1147 0.2737 0.3257 0.1933 0.0572 0.0084
P u1 = 2 u2 = 6 u3 = 10 u4 = 14 u5 = 18 u6 = 22
x12 0.1522 0.3051 0.3051 0.1522 0.0378 0.0046
x13 0.0828 0.2351 0.3328 0.2351 0.0828 0.0145 p0 0.6743 0.6672 0.6672 0.6949 1 0.6949
x14 0.2351 0.3328 0.2351 0.0828 0.0145 0.0013 p1 0.6743 0.6672 0.6743 1 0.3051 1
x15 0.1147 0.2737 0.3257 0.1933 0.0572 0.0084 p2 0.6743 0.6672 0.6743 0.2737 0.1522 0.0378
x16 0.0572 0.1933 0.3257 0.2737 0.1147 0.0239 p3 1 0.6672 0.6743 0.2351 0.1147 0.0239
x17 0.0003 0.0046 0.0378 0.1522 0.3051 0.3051 p4 0.3051 0.6672 0.6949 0.1933 0.0828 0.0145
x18 0.0378 0.1522 0.3051 0.3051 0.1522 0.0378 p5 0.3051 0.6743 1 0.1933 0.0572 0.0084
x19 0.1933 0.3257 0.2737 0.1147 0.0239 0.0025 p6 0.2737 0.6743 0.3051 0.1522 0.0572 0.0084
x20 0.1933 0.3257 0.2737 0.1147 0.0239 0.0025 p7 0.2351 0.6743 0.2737 0.1522 0.0378 0.0046
p8 0.2351 0.6949 0.2737 0.1147 0.0239 0.0025
p9 0.2351 0.6949 0.2351 0.1147 0.0239 0.0025
p10 0.2351 1 0.2351 0.0828 0.0145 0.0013
p11 0.2351 0.3051 0.2351 0.0828 0.0145 0.0013
Table 6
p12 0.1933 0.2737 0.2351 0.0828 0.0145 0.0013
Leaving information q
ij . p13 0.1933 0.2737 0.2351 0.0828 0.0145 0.0013
X1 u1 = 2 u2 = 6 u3 = 10 u4 = 14 u5 = 18 u6 = 22 p14 0.1522 0.2737 0.1933 0.0828 0.0145 0.0013
p15 0.1147 0.2737 0.1522 0.0572 0.0084 0.0006
x1 0.6743 0 0 0 0 0 p16 0.1147 0.2737 0.1522 0.0378 0.0046 0.0003
x2 0.6743 0 0 0 0 0 p17 0.0828 0.2351 0.1147 0.0378 0.0046 0.0003
x3 0 0.6672 0 0 0 0 p18 0.0572 0.1933 0.1147 0.0239 0.0025 0.0001
x4 0 0.6949 0 0 0 0 p19 0.0378 0.1522 0.1147 0.0239 0.0025 0.0001
x5 0 0.6672 0 0 0 0 p20 0.0003 0.0046 0.0378 0.0239 0.0025 0.0001
x6 0.6743 0 0 0 0 0
x7 0 0.6949 0 0 0 0
x8 0 0.6672 0 0 0 0
x9 0 0.6743 0 0 0 0 Table 9
x10 0 0.6672 0 0 0 0 The weights of probabilities in each interval.
x11 0 0 0.6743 0 0 0
x12 0 0 0.6949 0 0 0 w u1 = 2 u2 = 6 u3 = 10 u4 = 14 u5 = 18 u6 = 22
x13 0 0 0.6672 0 0 0 p0 0.3372 0.3336 0.3336 0.3474 0.8474 0.3474
x14 0 0.6672 0 0 0 0 p1 0 0 0.0036 0.5157 0.0765 0.6337
x15 0 0 0.6743 0 0 0 p2 0 0 0 0.0193 0.0187 0.0069
x16 0 0 0.6743 0 0 0 p3 0.5103 0 0 0.0209 0.016 0.0047
x17 0 0 0 0 0 0.6949 p4 0 0 0.0103 0 0.0128 0.003
x18 0 0 0 0.6949 0 0 p5 0.0157 0.0036 0.5 0.0206 0 0
x19 0 0.6743 0 0 0 0 p6 0.0193 0 0.0157 0 0.0097 0.0019
x20 0 0.6743 0 0 0 0 p7 0 0 0 0.0187 0.0069 0.0011
p8 0 0.0103 0.0193 0 0 0
p9 0 0 0 0.016 0.0047 0.0006
p10 0 0.5 0 0 0 0
p11 0.0209 0.0157 0 0 0 0
Table 7 p12 0 0 0 0 0 0
Joining information q
ij . p13 0.0206 0 0.0209 0 0 0
p14 0.0187 0 0.0206 0.0128 0.003 0.0003
X1 u1 = 2 u2 = 6 u3 = 10 u4 = 14 u5 = 18 u6 = 22
p15 0 0 0 0.0097 0.0019 0.0002
x1 0 0.2737 0.1147 0.0239 0.0025 0.0001 p16 0.016 0.0193 0.0187 0 0 0
x2 0 0.2737 0.1147 0.0239 0.0025 0.0001 p17 0.0128 0.0209 0 0.0069 0.0011 0.0001
x3 0.2351 0 0.2351 0.0828 0.0145 0.0013 p18 0.0097 0.0206 0 0 0 0
x4 0.3051 0 0.1522 0.0378 0.0046 0.0003 p19 0.0187 0.0738 0.0385 0 0 0
x5 0.2351 0 0.2351 0.0828 0.0145 0.0013 p20 0.0001 0.0023 0.0189 0.012 0.0012 0.0001
x6 0 0.2737 0.1147 0.0239 0.0025 0.0001
x7 0.3051 0 0.1522 0.0378 0.0046 0.0003
x8 0.2351 0 0.2351 0.0828 0.0145 0.0013
x9 0.2737 0 0.1933 0.0572 0.0084 0.0006
Table 10
x10 0.2351 0 0.2351 0.0828 0.0145 0.0013
The risk results of analyzing the times of oods based on DIOSM.
x11 0.1147 0.2737 0 0.1933 0.0572 0.0084
x12 0.1522 0.3051 0 0.1522 0.0378 0.0046 Name Interval Risk Expectation Probability Risk
x13 0.0828 0.2351 0 0.2351 0.0828 0.0145 Ij level value estimation
x14 0.2351 0 0.2351 0.0828 0.0145 0.0013
I1 [0, 4) 2 0.1746 0.185 1.0000
x15 0.1147 0.2737 0 0.1933 0.0572 0.0084
I2 [4, 8) 6 0.3878 0.411 0.8150
x16 0.0572 0.1933 0 0.2737 0.1147 0.0239
I3 [8, 12) 10 0.2381 0.2523 0.4039
x17 0.0003 0.0046 0.0378 0.1522 0.3051 0
I4 [12, 16) 14 0.0838 0.0888 0.1516
x18 0.0378 0.1522 0.3051 0 0.1522 0.0378
I5 [16, 20) 18 0.0238 0.0252 0.0627
x19 0.1933 0 0.2737 0.1147 0.0239 0.0025
I6 [20, 24) 22 0.0354 0.0375 0.0375
x20 0.1933 0 0.2737 0.1147 0.0239 0.0025
6218 Q. Zou et al. / Expert Systems with Applications 39 (2012) 62136220

Table 11 p14 p 0:6949=0 1=0:05 0:2737=0:1 0:2351=0:15


Fuzzy risk of the number of deaths represented by the PPD.
0:1933=0:2 0:1933=0:25 0:1522=0:3
P u1 = 404 u2 = 1212 u3 = 2020 u4 = 2828 u5 = 3636 u6 = 4444
0:1522=0:35 0:1147=0:4 0:1147=0:45
p0 0.6672 0.6672 0.6716 1 1 0.6738
p1 0.6672 0.6673 0.6946 0.2023 0.2981 0.6885 0:0828=0:5 0:0828=0:55 0:0828=0:6
p2 0.6673 0.6674 1 0.1526 0.2723 1
p3 0.6675 0.6698 0.298 0.1423 0.0621 0.0095 0:0828=0:65 0:0828=0:7 0:0572=0:75
p4 0.6676 0.673 0.2419 0.1421 0.0379 0.0047 0:0378=0:8 0:0378=0:85 0:0239=0:9
p5 0.6686 0.6884 0.2389 0.1133 0.0337 0.004
p6 0.6704 1 0.2296 0.0877 0.0158 0.0014 0:0239=0:95 0:0239=1
p7 0.6725 0.3049 0.21 0.0855 0.0152 0.0013
p8 0.6738 0.2833 0.1975 0.079 0.0135 0.0011
p9 0.6788 0.2687 0.1246 0.0665 0.0105 0.0008 In the form a/b, a is the possibility while b is the possibility. This is
p10 1 0.266 0.1099 0.0594 0.0089 0.0007
means, the probability for u4 = 14 is the largest possibility value 1
p11 0.2702 0.2617 0.1036 0.0273 0.003 0.0002
p12 0.2591 0.2531 0.0964 0.0224 0.0023 0.0001
when p = 0.05, but other values should not be ignored. when
p13 0.2405 0.2443 0.0895 0.0204 0.002 0.0001 p = 0.10, the probability is 0.2737. Nevertheless, even if p = 0 for it
p14 0.2312 0.2403 0.0865 0.0182 0.0017 0.0001 would not happen more having the possibility value 0.6949.
p15 0.2282 0.2387 0.0854 0.0163 0.0015 0.0001 Step 5: For the PPD, calculate the weights for each interval as
p16 0.1625 0.2347 0.0825 0.0155 0.0014 0.0001
shown in Table 9, and then calculate the expected values
p17 0.1518 0.2245 0.0756 0.0152 0.0013 0.0001
p18 0.1074 0.1948 0.0579 0.0144 0.0012 0.0001 of probabilities in every interval uj.
p19 0.0002 0.004 0.0338 0.0126 0.001 0 Step 6: Using Eqs. (17) and (18) respectively to calculate the prob-
p20 0.0001 0.0024 0.0235 0.0086 0.0006 0 ability p(uj) and risk estimation value P(uj), and the results
of expected values of probabilities, p(uj) and P(uj) are shown
in Table 10. The column separately shows the name of the
intervals, the middle point also called risk level of each
Table 12
The risk results of analyzing the number of deaths based on DIOSM. interval, expectation of probabilities dropping in each inter-
val, the probability value as well as the risk estimation.
Name Interval Ij Risk Expectation Probability Risk
level value estimation
Finally calculate the total fuzzy expected value E with Eqs.
I1 [0, 808) 404 0.3738 0.4187 1.0000
(19)(21).
I2 [808, 1616) 1212 0.2747 0.3078 0.5813
I3 [1616, 2424) 2020 0.1255 0.1406 0.2735
X
m
I4 [2424, 3232) 2828 0.0325 0.0364 0.1330 E uj  px 2 Ij
I5 [3232, 4040) 3636 0.0196 0.022 0.0965
j1
I6 [4040, 4848) 4444 0.0665 0.0746 0.0746
2  0:2124 6  0:2798 10  0:2351 14  0:1662
tively, and the results of Q and Q are shown in Tables 6
18  0:0755 22  0:0310 8:82  9
j j
and 7. It means that there are about 8 or 9 oods every year, and more
Step 4: Using Eqs. (4)(6) to get the obtain the fuzzy risk calcu- likely to 9. And from Table 10, it tell us there are indeed 2 oods
lated by the PPD, as shown in Table 8. From the Table 5 every year, the probability of 48 oods is largest of all to 0.411,
above, we could not obtain the exact estimation for prob- the probability of the time of oods above 14 is 0.1516. And the
ability, but a series values. phenomenon is seldom when the time is beyond 18 with the prob-
Here, the universe of discourse of probability is ability 0.0627, for 1/0.0627 = 15.95  16, it means the time of
P fp0 ; p1 ; p2 ; . . . ; p19 ; p20 g = {0, 0.05, 0.10, . . . , 0.90, 0.95, 1}. oods over 18 would occur every 16 years. Finally, the phenome-
And to u4 = 14, non is rare when the time above 22 with the probability 0.0375,

Table 13
Fuzzy risk of economic losses represented by a the PPD.

P u1 = 3142.8 u2 = 9428.4 u3 = 15714 u4 = 22000 u5 = 28285.2 u6 = 34570.8


p0 0.6686 0.6673 0.6676 0.6939 1 0.6898
p1 0.6688 0.6727 1 1 0.2996 1
p2 0.6711 0.6775 0.3042 0.225 0.1537 0.0384
p3 0.6733 0.688 0.2306 0.1443 0.0759 0.0127
p4 0.6738 1 0.1983 0.0796 0.0137 0.0012
p5 0.6744 0.3017 0.1848 0.0599 0.009 0.0007
p6 0.6762 0.2806 0.1633 0.0527 0.0075 0.0005
p7 0.6774 0.2741 0.1472 0.0425 0.0055 0.0004
p8 0.6828 0.2725 0.1217 0.0358 0.0043 0.0003
p9 0.684 0.245 0.115 0.0262 0.0028 0.0001
p10 0.6869 0.2166 0.1135 0.024 0.0025 0.0001
p11 0.6898 0.2153 0.0705 0.0235 0.0024 0.0001
p12 0.6916 0.2044 0.0697 0.0114 0.0009 0
p13 1 0.1964 0.0633 0.0112 0.0009 0
p14 0.2974 0.188 0.0588 0.0097 0.0007 0
p15 0.2808 0.1729 0.0544 0.0087 0.0006 0
p16 0.2694 0.1705 0.0469 0.0078 0.0006 0
p17 0.2396 0.1652 0.0458 0.0063 0.0004 0
p18 0.09 0.1602 0.0434 0.0061 0.0004 0
p19 0.0372 0.1507 0.0412 0.0057 0.0004 0
p20 0.0002 0.0041 0.0346 0.0053 0.0003 0
Q. Zou et al. / Expert Systems with Applications 39 (2012) 62136220 6219

Table 14
The risk results of analyzing economic losses based on DIOSM.

Name Interval Ij Risk level Expectation Probability value Risk estimation


I1 [0, 6285.6) 3142.8 0.4599 0.5350 1.0000
I2 [6285.6, 12571.2) 9428.4 0.2128 0.2476 0.4650
I3 [12571.2, 18856.8) 15714 0.0860 0.1000 0.2175
I4 [18856.8, 25142.4) 22000 0.0523 0.0608 0.1175
I5 [25142.4, 31428) 28285.2 0.0145 0.0169 0.0566
I6 [31428, 37713.6) 34570.8 0.0341 0.0397 0.0397

for 1/0.0375 = 26.67  27, also providing that the time of oods Through the application of DIOSM, we analyze the fuzzy risk of
over 22 would occur every 2627 years. Last but not the least, ood disasters and calculate the ood risk quantitatively in China
we are able to get some other information from Tables 810. by considering the time of oods, the number of the deaths and
the economic losses each year from 1990 to 2009. It is hoped that
3.2. Fuzzy risk analysis using the number of deaths based on DIOSM the results would provide the government, engineers, analysts,
decision-makers and local authorities with a more suitable and
And for the number of deaths caused by oods, the process of invaluable guide and overview on ooding, which is helpful for
DIOSM to calculate the fuzzy risk is almost the same. By the way, them to outline the policy and practice of managing ood risk
taking consideration the number should be integer, we obtain clearly. Moreover, the questions on how to respond, treat and de-
D = 808, and get the corresponding intervals. For economy of fend when we encounter with ooding can be solved, such as the
space, we shall not go into detail, and we put out the results shown use of ood warning system, implementation of temporary ood
in Tables 11 and 12. And the total fuzzy expected value for the defenses, measures to protect valuable and mobilization of emer-
number of deaths is 1648. It means that there are about 1648 gency services. Nonetheless, if more samples can be collected,
deaths every year. And from Table 12, it shows us that the proba- the results will be more objective and accurate. There is no doubt
bility of 0808 deaths is largest of all to 0.4187, the probability of that fuzzy risk analysis will be a key feature in the ood risk man-
the number of death above 2020 is 0.2735, and the phenomenon is agement over the years all over the world. And DIOSM based on
seldom when the number of deaths is beyond 3636 and rare when information diffusion theory will ensure that we can identify, eval-
the number of deaths above 4444 with the probability 0.0746. In uate quantitatively, control, and mitigate risks associated with hu-
addition, we are able to get some other information from Tables man activities, thus manage future ood risk in China in a most
11 and 12. effective way.

3.3. Fuzzy risk analysis using the economic losses based on DIOSM
Acknowledgements
At last, as for the economic losses caused by oods, the proce-
This work is supported by the National Key Basic Research Pro-
dure based on DIOSM to calculate the fuzzy risk is similar. With
gram of China (973 Program, Grant No. 2007CB714107), the Na-
limited space, we just give out the results in Tables 13 and 14.
tional Science & Technology Pillar Program of China (Grand No.
And the total fuzzy expected value for the economic losses caused
2008BAB29B08), and the Special Research Foundation for the Pub-
by oods is 11106.99 thousand dollars every year. And from Ta-
lic Welfare Industry of the Ministry of Water Resources (Grant Nos.
ble 14, the probability of 06285.6 million is largest of all to
201001080 and 200901010).
0.5341, the probability of economic losses above 15714 is 0.2735,
and the phenomenon is seldom when economic losses are beyond
28285.2 thousand dollars and rare when economic losses above References
34570.8 thousand dollars with the probability 0.0397.
Bardossy, A., Bogardi, I., & Duckstein, L. (1991). Fuzzy set and probabilistic
techniques for health-risk analysis. Applied Mathematics and Computation,
4. Discussions and conclusions 45(3), 241268.
Cheng, S. R., Lin, B. S., Hsu, B. M., & Shu, M. H. (2009). Fault-tree analysis for liqueed
In this work, the standard interior-outer-set model (IOSM) natural gas terminal emergency shutdown system. Expert Systems with
Applications, 36(9), 1191811924.
based on the information diffusion theory is introduced rstly. Fleming, George (2002). Flood risk management: Learning to live with rivers. Thomas
Considering its three deciencies, we give the measurement to se- Telford Publishing.
lect the domain and step length of the universe, as well as apply Feng, L. H., Hong, W. H., & Wan, Z. (2010). The application of fuzzy risk in
researching ood disasters. Natural Hazards, 53, 413423.
the normal diffusion function to replace the one dimensional linear Fisher, B. (2006). Fuzzy approaches to environmental decisions: Application to air
information-distribution. Then the diffusion interior-outer-set quality. Environmental Science & Policy, 9(1), 2231.
model (DIOSM) is put forward, which has the advantages of choos- Fu, G. T. (2008). A fuzzy optimization method for multicriteria decision making: An
application to reservoir ood control operation. Expert Systems with Applications,
ing the domain and step length reasonably, changing smoothly 34(1), 145149.
with gradient descent from 0 to 1 in the PPD, and making the val- Huang, C. F., & Ruan, D. (2008). Fuzzy risks and an updating algorithm with new
ues dispersive as well as seldom equal to 0. Therefore, its three observations. Risk Analysis, 28(3), 681694.
Huang, C. F. (1996). Fuzzy risk assessment of urban natural hazards. Fuzzy Sets and
advantages can be easily seen and demonstrated from the PPDs Systems, 83(2), 271282.
based on DIOSM shown in Tables 8, 11 and 13. Huang, C. F. (1997). Principle of information diffusion. Fuzzy Sets and Systems, 91(1),
The number of the sample points is only 20, with the statistical 6990.
Huang, C. F. (2004). A demonstration of reliability of the interior-outer-set model.
data from the year 1990 to 2009; nobody could precisely evaluate
International Journal of General Systems, 33(23), 205222.
the underlying probability function. With the conventional models, Huang, C. F., Zong, T., & Chen, T. F. (2007). Four models to calculate a fuzzy
we were unable to get a correct assumption. However, the DIOSM probability distribution with a small sample. International Journal of Information
is capable of calculating the possibilityprobability distribution Technology and Decision Making, 6(4), 611623.
Huang, C. F., & Moraga, C. (2002). A fuzzy risk model and its matrix algorithm.
with a small sample and can give the better multi-value results International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems,
as much as possible to support ood risk management. 10(4), 347362.
6220 Q. Zou et al. / Expert Systems with Applications 39 (2012) 62136220

International Federation of Red Cross and Red Crescent Societies. (1998). World Rainer, R. K., Snyder, C. A., & Carr, H. H. (1991). Risk analysis for information
disaster report. Oxford: Oxford University Press. technology. Journal of Management Information Systems, 8(1), 129147.
Karimi, I., & Hllermeier, E. (2007). Risk assessment system of natural hazards: A Reyna, V. F., & Brainerd, C. J. (2008). Numeracy, ratio bias, and denominator neglect
new approach based on fuzzy probability. Fuzzy Sets and Systems, 158(9), in judgments of risk and probability. Learning and Individual Differences, 18(1),
987999. 89107.
Lee, L. W., & Chen, S. M. (2008). Fuzzy risk analysis based on fuzzy numbers with Shen, Q., & Zhao, R. Q. (2010). Risk assessment of serious crime with fuzzy random
different shapes and different deviations. Expert Systems with Application, 34(4), theory. Information Sciences, 180(22), 44014411.
27632771. Wang, Z. Y., & Plate, E. J. (2002). Recent ood disasters in China. ICE Proceedings,
Li, X. P. (1994). Probability theory foundation. Higher Education Publishing Company. Water & Maritime Engineering, 154, 77188.
in Chinese. Wei, Y. M., Jin, J. L., Yang, C. J., et al. (2002). Theory of risk management of ood
Liu, B., & Liu, Y. K. (2002). Expected value of fuzzy variable and fuzzy expected value disaster. Beijing: Science. in Chinese.
models. IEEE Transactions on Fuzzy Systems, 10(4), 445450. White, D. (1995). Application of systems thinking to risk management: A review of
Misra, K. B., & Weber, G. G. (1990). Use of fuzzy set theory for level-I studies in the literature. Management Decision, 3(10), 3545.
probabilistic risk assessment. Fuzzy Sets and Systems, 37(2), 139160. Warmerdama, J. M., & Jacobsa, T. L. (1994). Fuzzy set approach to routing and siting
Mikhailov, V. N., Morozov, V. N., Cheroy, N. I., et al. (2008). Extreme ood on the hazardous waste operations. Information Sciences-Applications, 2(1), 114.
Danube River in 2006. Russian Meteorology and Hydrology, 33(1), 4854. Zhang, J. Q., Zhou, C. H., Xu, K. Q., & Watanabe, M. (2002). Flood disaster monitoring
Ngai, E., & Wat, F. (2005). Fuzzy decision support system for risk analysis in e- and evaluation in China. Global Environmental Change Part B: Environmental
commerce development. Decision Support Systems, 40(2), 235255. Hazards, 4(23), 3343.
Parker, D. J., (2000). Introduction to oods and ood management. In: Parker, D. J. Zadeh, L. A. (1978). Fuzzy sets as a basis for a theory of possibility. Fuzzy Sets and
(Ed.), Floods. London & New York (Vol. I, pp. 339). Systems, 328.

You might also like