You are on page 1of 10

Expert Systems With Applications 42 (2015) 9379–9388

Contents lists available at ScienceDirect

Expert Systems With Applications


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

A prospect theory-based interval dynamic reference point method for


emergency decision making
Liang Wang , Zi-Xin Zhang , Ying-Ming Wang∗
School of Economics and Management, Fuzhou University, Fuzhou 350116, PR China

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

Keywords: Urgent or critical situations, such as terrorist attacks and natural disasters, often require decision makers
Emergency decision making (DMs) to take crucial decisions. Emergency decision making (EDM) problems have become a very active re-
Emergency response
search field in recent years. The existing studies focus mainly on the information inadequacy or incomplete
Prospect theory
information in emergencies, and selecting ideal emergency alternatives, neglecting the psychological behav-
TOPSIS
Interval value ior of DMs under emergencies. Few studies consider a DM’s psychological behavior, although there is some
Dynamic reference point focus on the dynamic features of emergency events and limited DM judgments under risk and uncertainty.
Motivated by such problems, this study proposes a prospect theory-based interval dynamic reference point
method for EDM. The technique for order preference by similarity to ideal solution (TOPSIS) method is a pop-
ular decision technique, as decision makers are always bounded rational under risk and uncertainty and their
psychological behavior plays an important role in EDM. However, the existing TOPSIS methods are seldom
concerned with this issue. Based on such a problem, this study proposes a TOPSIS method with an interval
reference point, which considers the DM’s psychological behavior. Two examples are presented to illustrate
the feasibility and validity of the proposed methods for solving EDM problems in the real world. Based on
the final rankings of alternatives in the examples, each of our methods validates the other and matches the
actual EDM.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction theory, including risk analysis to select a strategy for protecting


populations after simulated nuclear accidents. Mendonça, Rush, and
With the rapid development of technology, economy, and society, Wallace (20 0 0) considered that two crucial factors for effective emer-
human beings face all kinds of social problems, which are increas- gency response were the speed with which the response strategy
ingly diversified and complicated. In particular, in recent years, var- was implemented and the quality of the expert knowledge on which
ious emergency events, such as air crashes, earthquakes, hurricanes, the response was based; thus, they proposed a method for using
and floods, have had severely negative impacts on human life and communication and computing technologies to elicit and aggregate
social development. When such a devastating incident occurs, how the knowledge of multiple, geographically separate experts that im-
to take effective and appropriate measures to control for the esca- proved emergency response ability. Millar, Joseph, and Mobolurin
lation and deterioration of the situation is of practical significance. (2002) proposed a mathematical programming model to help the
Therefore, an importance research topic for emergency management decision maker (DM) select a maximum value of the recovery ca-
is how to undertake a response timeously and effectively, or how pability of a recover strategy. Sun, Ma, and Zhao (2013) proposed
to select a desirable alternative in the earlier stage of an emergency a method based on fuzzy rough set theory over two universes to
event, and many researchers have studied such problems. predict emergency material demand. Yang and Xu (2012) estab-
Many studies have approached this topic from different angles. lished an engineering model based on sequential games in order to
For example, Rainabralimam, Sreenivasulu, and Mallikaijunan (1996) establish the optimal relief plan and provide support for decision
provided an overview of release scenarios and affected areas, re- making when an emergency occurs. Yu and Lai (2011) proposed a
sponses, emergency procedures, and so on. Hämäläinen, Lindstedt, distance-based group decision-making (GDM) method to solve un-
and Sinkko (20 0 0) developed a method for multi-attribute utility conventional multi-criteria emergency decision-making problems.
De Maio, Fenza, and Gaeta (2011) presented an approach based on

Corresponding author. Tel.: +86 591 22866677; fax: +86 591 22866677. fuzzy cognitive maps to support knowledge processing and resource
E-mail addresses: wangliangg322@hotmail.com (L. Wang ), zhangzixin768@ discovery according to emergency features. Wu and Zhao (2010)
outlook.com (Z.-X. Zhang ), msymwang@hotmail.com (Y.-M. Wang). proposed a model of emergency decision making (EDM) based on

http://dx.doi.org/10.1016/j.eswa.2015.07.056
0957-4174/© 2015 Elsevier Ltd. All rights reserved.
9380 L. Wang et al. / Expert Systems With Applications 42 (2015) 9379–9388

knowledge supply and demand. Zhao, Jin, and Shen (2007) proposed information features of emergency events. Furthermore, existing
a case-based reasoning decision-making method to deal with infor- studies (Fan et al., 2012; Kahneman & Tversky, 1979; Liu et al., 2014;
mation inadequacy, as well as uncertainty and dynamic trends in cri- Tversky & Kahneman, 1992; Wang & Wang, 2013) used crisp num-
sis management. Körte (2003) presented a contingent risk and deci- bers as the reference points of PT; however, in the real world, it is
sion analysis method to provide decision support for decisions un- difficult for DMs to estimate the damage, loss, or cost of the emer-
der variable circumstances and short available time scales. Levy and gency event using a crisp number as the reference point under inade-
Taji (2007) proposed a group analytic network process to support quate/incomplete information; interval values are often more suit-
hazard planning and emergency management under incomplete in- able for estimation. Decision making problems in emergencies are
formation. Ergu, Kou, and Shi (2011) proposed a simple consistency usually risky, uncertain, and dynamic evolutionary, and, as is well
test and inconsistent element identification approach to make ana- known, DMs are always bounded rational under risk and uncertainty.
lytic network process (ANP) easier and suitable to deal with decision- Motivated by such problems, there is, therefore, a need to develop a
making problems in emergencies. Fu, Wu, and Tang (2012) proposed decision analysis method for EDM considering the DM’s psychologi-
an intelligent GDM methodology to solve unconventional emergency cal behavior and the dynamic and incomplete information features of
management problems aimed at improving decision accuracy, en- emergency events.
hancing decision transparency, and increasing decision effectiveness. Since Kahneman and Tversky (1979) firstly proposed PT, behav-
Xiong and Li (2011) proposed a local GDM model based on variable ioral decision making theories have developed rapidly, for example,
precision rough set to deal with uncertainty, which is the result of regret theory (Bell, 1982; Loomes and Sugden, 1982), disappointment
lack of information in emergency GDM. Fan, Liu, and Shen (2012) theory (Bell, 1985; Loomes & Sugden, 1986), cumulative PT (Tversky
proposed a risk decision analysis method for emergency response & Kahneman, 1992), third-generation PT (Schmidt, Starmer, & Sug-
based on prospect theory (PT). Wang and Wang (2013) proposed a den, 2008). Among these theories, PT is regarded as the most in-
PT-based dynamic reference point for emergency decision making. fluential theory of decision making under uncertainty (Abdellaoui,
With respect to the multiple states and different outcomes resulting Bleichrodt, & Paraschiv, 2007; Bleichrodt, Schmidt, & Zank, 2009;
from the evolution of emergency events, Liu, Fan, and Yuan (2013) Schmidt & Zank, 2008; Schmidt et al., 2008). This is because PT mod-
proposed a multiple attribute decision making method for risk. Liu, els two major deviations from expected utility: nonlinear decision
Fan, and Yuan (2014) proposed a risk decision-analysis method based weighting and loss aversion, that is, the tendency that people treat
on cumulative PT to solve decision-making problems in emergency outcomes as deviations from a reference point and are more sensitive
responses. Liu et al. (2014) proposed a method based on fault tree to losses than gains of the same magnitude. Both nonlinear decision
analysis to solve decision-making problems in emergency responses. weighting and loss aversion are documented widely in the empiri-
Zhou, Wang, and Zhuo (2014) proposed a novel planning approach to cal literature (Abdellaoui et al., 2007; Bleichrodt et al., 2009; Booij
coordinate for solutions among agents in a multi-agent system. Ergu, and Kuilen, 2009; Wu & Markle, 2008). PT describes the DM’s be-
Zhang, Guo, and Yang (2014) provided experimental consistency sim- havioral characteristics well and provides the calculation formulas
ulations for a Hadamard product induced bias matrix (HPIBM), which for gains, losses, and values. Since the formulas have the character-
can be used to identify the most inconsistent entry in an emergency istics of a simple computation process and clear logic, PT has been
decision matrix by randomly generating decision matrixes with or- used widely to solve various decision making problems considering
ders three to seven. In addition, they proposed a novel HPIBM for DMs’ psychological behavior. Therefore, more attention is needed on
estimating the missing judgments in an emergency decision matrix how to incorporate PT into decision analysis in EDM. In this study,
by constructing an optimization problem while improving the con- we propose a PT-based interval dynamic reference point method for
sistency ratio. With regard to the risk characteristics of decision mak- EDM.
ing by large groups in emergencies and difficulties forming decision The technique for order preference by similarity to ideal solution
schemes with low risk, Xu, Liang, Chen, and Zhou (2015) provided (TOPSIS) method, proposed by Hwang and Yoon (1981) in 1981, is
methods of risk measure and elimination for emergency decisions a popular approach for multiple criteria decision making (MCDM).
to construct coordination frameworks and risk elimination coordi- Since Hwang and Yoon (1981) proposed TOPSIS, it has been used
nation mechanisms for emergency decisions. Mo, Duan, Shen, and and studied widely in the literature (Abo-Sinna & Amer, 2005; Chen,
Wang (2015) developed an interval-parameter two-stage stochastic 2011, 2015; Chen & Tsao, 2008; Cheng, Chan, & Huang, 2003; Chu,
integer programming model for urban water resource planning un- 20 02a, 20 02b; Chu & Liu, 20 03, 20 09; Dymova, Sevastjanov, & Tikho-
der uncertainty. Ju, Wang, and You (2015) proposed a new frame- nenko, 2013a, 2015; Krohling & Campanharo, 2011; Lai, Liu, & Hwang,
work combining the ANP method, the decision-making trial and eval- 1994; Mahdevari, Shahriar, & Esfahanipour, 2014; Mokhtarian, 2014;
uation laboratory technique, and the 2-tuple linguistic technique Rashid,2014; Wang & Elhag, 2006; Ye, 2010). Since the DM’s judg-
for order preference by similarity to an ideal solution method in ments are often vague under risk and uncertain situations, crisp
order to solve the emergency alternative evaluation and selection numbers may not be determined precisely; therefore, fuzzy values
problem. or interval values on TOPSIS methods have been proposed and dis-
The abovementioned literature review shows that existing studies cussed to solve problems under a fuzzy environment (Abo-Sinna &
mainly are concerned with approaches that deal with incomplete or Amer, 2005; Chen, 2012, 2015; Chen & Lee, 2010; Chen & Tsao, 2008;
inadequate information in EDM situations (Ergu et al., 2014; Fu et al., Cheng et al., 2003; Chu, 2002a, 2002b; Chu & Liu, 2003, 2009; Deepa
2012; Levy & Taji, 2007; Xiong & Li, 2011; Zhao et al.,2007) or that and Sanjay, 2015; Dymova et al., 2013b, 2015; Jahanshahloo, Lotfi, &
select the ideal emergency alternatives (De Maio et al., 2011; Ergu Davoodi, 2009; Krohling & Campanharo, 2011; Lai et al., 1994; Tsaur,
et al., 2011; Hämäläinen et al., 20 0 0; Ju et al., 2015; Körte, 2003; Liu 2011; Wang & Elhag, 2006; Yue,2014). However, the existing inter-
et al., 2014; Mendonça et al., 20 0 0; Millar et al., 2002; Sun et al., 2013; val TOPSIS methods are rarely concerned about the interval reference
Xu et al., 2015; Yang & Xu, 2012; Zhou et al., 2014;). As is generally point, which reflects the DM’s psychological behavior under risk and
known, the psychological behavior of DMs plays an important role uncertainty. According to risk propensity, it has been commonly ob-
in emergency response. However, neither strand of literature takes served that DMs differ in willingness to overestimate the probabil-
the decision maker’s (DM’s) psychological behavior into account and ity of gain or loss under emergencies. As mentioned, since the DMs
there are few studies in EDM that consider the DM’s psychological be- are always bounded rational under risk and uncertainty, a DM’s psy-
havior (Fan et al., 2012; Liu Fan & Zhang, 2014; Wang & Wang, 2013). chological behavior plays an important role in risky and uncertain
Nevertheless, the literature considering the DM’s psychological be- situations. How to incorporate interval reference points into the TOP-
havior does not consider the dynamic and inadequate/incomplete SIS deserves to be mentioned. Motivated by such a need, this study
L. Wang et al. / Expert Systems With Applications 42 (2015) 9379–9388 9381

discusses the TOPSIS method with interval reference points under Value
emergencies. v( x)
In general, the criteria can be classified into two types: benefit
and cost (Ma, Fan, & Huang, 1999). The benefit criterion means that a
higher value is better; while the cost criterion means that a lower
v( x0 )
value is better. This characteristic is also applied to the reference
point of PT. For the cost type, based on the idea of PT, if the attribute
value of each alternative is above the reference point, the part in ex- − x0
Losses Gains
cess can be regarded as the “loss” to the DM; if the attribute value of 0 x0 x
each alternative is below the reference point, the rest can be regarded
as the “gain” to the DM. In fact, the DMs have different psychological
mirrors for gains and losses, which are more sensitive to losses than
to gains of the same magnitude. The gains and losses of all alterna- v(− x0 )
tives are constructed into a matrix, known as the gain loss matrix
(GLM). According to the GLM, by using the value function of PT, the
value matrix (VM) is constructed. Since different attributes are usu-
ally incommensurate, the VM needs to be normalized into a corre- Fig. 1. An S-shapes value function of PT.

sponding element in matrix VM so as to transform it into comparable


values. Furthermore, by using the simple additive weighting method,
this study takes the values as follows: α = 0.89, β = 0.92, λ = 2.25
the overall prospect value of each alternative is obtained, and then,
(Tversky & Kahneman, 1992).
a ranking order of all alternatives is determined. The proposed TOP-
SIS method with interval reference point is similar to the PT-based
2.2. The TOPSIS method
interval dynamic reference point method in obtaining the GLM. After
obtaining the GLM, the value of each alternative is calculated by the
The TOPSIS method was proposed initially by Hwang and Yoon
classical TOPSIS method, and then, a ranking order of all alternatives
(1981). Since then, many studies on TOPSIS have been conducted
is determined. Finally, based on the ranking order, the ideal alterna-
(Abo-Sinna & Amer, 2005; Chen, 2011; Chu & Liu, 2003, 2009; Chen &
tive is chosen for emergency response.
Tsao, 2008; Cheng et al., 2003; Chu, 2002a, 2002b; Krohling & Cam-
The remainder of this paper is organized as follows. Section 2
panharo, 2011; Lai et al., 1994; Wang & Elhag, 2006)
briefly introduces the PT and TOPSIS methods. The proposed method
Suppose an MCDM problem has n alternatives,S1 , . . . , Sn ,and m
is described in Section 3. Section 4 discusses the different stages of
decision criteria/attributes, C1 , . . . , Cm . xi j indicates the ratings/values
emergency response and the adjustment of alternatives to the pro-
of the alternatives with respect to each criterion, which forms a de-
posed PT-based method considering the interval dynamic reference
cision matrix denoted by X = (xi j )n×m . Let W = (w1 , . . . , wm ) be the
point. In Section 5, two examples are provided to illustrate the use of
m
the proposed methods. Conclusions are offered in Section 6. relative weight vector about the criteria, satisfying w j = 1. In gen-
j=1
2. Prospect theory and TOPSIS method eral, the criteria can be classified into two types: benefit and cost (Ma
et al., 1999). The benefit criterion means that a higher value is better
2.1. Prospect theory
while the cost criterion means that a lower value is better. In addi-
Since Kahneman and Tversky (1979) first proposed PT, which is a tion, this characteristic is applied to the ideal solution (also called
descriptive model of decision making under conditions of risk, sev- positive ideal solution) and negative-ideal solution (also called anti-
eral studies on PT have been conducted (Abdellaoui et al., 2007; ideal solution). The ideal solution is one that maximizes the bene-
Bleichrodt et al., 2009; Booij and Van, 2009; Nwogugu, 2006; Tver- fit criteria/attributes and minimizes the cost criteria/attributes; on
sky & Kahneman, 1981, 1991, 1992; Schmidt et al., 2008; Schmidt & the other hand, the negative-ideal solution maximizes the cost crite-
Zank, 2008; 2012; Wakker, 2010; Wu & Markle, 2008). PT is described ria/attributes and minimizes the benefit criteria/attributes. The best
briefly below. alternative is the solution that is farthest from the negative-ideal so-
PT distinguishes two phases in the selection process: an early lution and closest to the ideal solution. The TOPSIS method is intro-
phase of editing and a subsequent phase of evaluation (Kahneman duced in the following steps:
& Tversky, 1979). In the editing phase, outcomes are expressed by Step 1: Normalize the decision matrix X = (xi j )n×m using the fol-
means of gains and losses from a reference point. In the evaluation lowing equation:
phase, the edited prospects are evaluated by a value function and a xi j
weighting function, and the prospect of highest value is selected. The ri j =  , i = 1, . . . , n; j = 1, . . . , m (2)

n
value function is defined on deviations from the reference point. The x2i j
value function used in PT is expressed in the form of a power law i=1

according to the following expression (Kahneman & Tversky, 1979). Normalized decision matrix X̄ = (ri j )n×m , where ri j is the normalized

xα , x≥0 criteria/attribute rating/value.
v(x) = (1) Step 2: Calculate the weighted normalized decision matrix V =
−λ( − x)β , x<0
(vi j )n×m
where x denotes the gains or losses; x ≥ 0 represents the gains and
x < 0 represents the losses. α and β are exponent parameters related
vi j = w j ri j , i = 1, . . . , n; j = 1, . . . , m (3)
to gains and losses, respectively, 0 ≤ α , β ≤ 1. λ is the risk aversion where w j is the relative weight of the jth criterion or attribute,
parameter, which represents the characteristic of steeper for losses 
m
and w j = 1.
than for gains, λ > 1. Fig. 1 shows a prospect value function with con- j=1
vex and concave S-shapes for losses and gains, respectively. The val- Step 3: Determine the ideal and negative-ideal solutions:
ues of α , β , and λ are determined through experiments (Abdellaoui
et al., 2007; Bleichrodt et al., 2009; Booij and Van, 2009; Tversky & S∗ = {v∗1 , . . . , v∗m }
Kahneman, 1992; Wakker, 2010). For simplifying the computation, = {( max vi j )| j ∈ b , ( min vi j )| j ∈ c } (4)
j j
9382 L. Wang et al. / Expert Systems With Applications 42 (2015) 9379–9388

Table 1
S − = {v−
1 m}
, . . . , v− Possible cases of positional relationship between R j and Ci j .

= {( min vi j )| j ∈ b , ( max vi j )| j ∈ c } (5) Cases Positional relationship between R j and Ci j


j j
Cij Rj
where b and c are the sets of benefit criteria/attributes and cost
criteria/attributes, respectively. CijL CijH R Lj RH
Case 1 CiHj < RLj j
Step 4: Calculate the Euclidean distances of each alternative from
the ideal solution and the negative-ideal solution, respectively: Rj Cij



m
RHj < CiLj R Lj RH CijL CijH
D∗i = (vi j − v∗j )2 , i = 1, . . . , n (6) Case 2 j

j=1 Cij
Rj


m
Case 3 CiLj < RLj ≤ CiHj < RHj CijL R Lj CijH RH
(vi j − v−j )2 ,
j
D−
i
= i = 1, . . . , n (7)
j=1 Cij
Rj

L L H
Step 5: Calculate the relative closeness RCi for each alternative R C R CijH
Case 4 RLj < CiLj ≤ RHj < CiHj j ij j

with respect to the ideal solution. The relative closeness of the al-
Cij
ternative Si with respect to S∗ is defined as: Rj
CijL RL CijH
D− CiLj < RLj < RHj < CiHj
Case 5 j RH
j

i
RCi = −, i = 1, . . . , n (8)
D∗i + Di Rj
Cij
Step 6: Rank the alternatives according to their relative closeness Case 6 RLj ≤ CiLj < CiHj ≤ RHj R L
j
C L
ij CijH RH
j

to the ideal solution. The bigger RCi is, the better is alternative Si . The
best alternative is the one with the greatest relative closeness to the
ideal solution.
According to Definition 1, the loss to the DM is given by
3. Proposed method CiHj
Li j = (x − RLj ) f (x)dx, i ∈ N, j ∈ M (11)
CiLj
This section mainly discusses the relationship between the inter-
val reference points and interval attribute values in an emergency. Obviously, by Eq. (9), Eq. (11) can be rewritten as
Thereafter, it provides the formulation and methodology for obtain-
ing the GLM and VM. Li j = 0.5(CiLj + CiHj ) − RLj , i ∈ N, j ∈ M (12)

Similarly, for case 2, since RHj < CiLj , the gain and loss to the DM are
3.1. Relationship between interval reference points and attribute values given by
 CiHj
Let M = {1, 2, . . . , m} and N = {1, 2, . . . , n}. Let R j = [RLj , RHj ], RHj > Gi j = (x − RHj ) f (x)dx
CiLj (13)
RLj be an interval reference point, where R j denotes the DM’s psy- = 0.5(CiHj + CiLj ) − RHj , i ∈ N, j ∈ M
chological reference point with respect to the jth attribute, j ∈ M.
Let Ci j = [CiLj , CiHj ], CiHj > CiLj be an interval attribute value, where Ci j Li j = 0, i ∈ N, j ∈ M (14)
denotes the jth attribute value with respect to the ith alternative,
i ∈ N, j ∈ M. Without loss of generality, we suppose that RLj ≥ 0 and respectively.
For case 3, let y be an arbitrary value in interval number [RLj , CiHj ].
CiLj ≥ 0. There are six possible cases between R j and Ci j in the number
Since the interval [RLj , CiHj ] ⊂ [RLj , RHj ], any possible value y in interval
axis (Fan, Zhang, & Chen, 2013), as shown in Table 1.
number [RLj , CiHj ] is equally acceptable to the DM. Thus, we need to
3.2. Calculation of gains and losses consider only the interval number [CiLj , RLj ]. Let x be an arbitrary value
in interval number [CiLj , RLj ], then, the gain and loss to the DM are given
For the relationship between R j and Ci j , as shown in Table 1, we by
provide the following definition.
Gi j = 0, i ∈ N, j ∈ M (15)
Definition 1. For interval attribute value Ci j , let x be an arbitrary  RLj
value in interval number [CiLj , CiHj ], regarded as a random variable with Li j = (x − RLj ) f (x)dx
CiLj (16)
uniform distribution. The probability density function of x is = 0.5(CiLj − RLj ), i ∈ N, j ∈ M
 1
, CiLj ≤ x ≤ CiHj respectively. When RLj = CiHj , the interval number [RLj , CiHj ] is a crisp
f (x) = CiHj − CiLj , i ∈ N, j ∈ M (9)
number, that is, y = RLj = CiHj , and the loss to the DM is equal to
0, otherwise
Eq. (11). By simplification, we can obtain the same result as Eq. (16).
 CiHj Similarly, for case 4, since RLj < CiLj ≤ RHj < CiHj , the gain and loss to
where f (x)dx = 1 and f (x) ≥ 0 for all x ∈ [CiLj , CiHj ].
CiLj the DM are
From Table 1, there are six cases, which are discussed as follows.
Gi j = 0.5(CiHj − RHj ), i ∈ N, j ∈ M (17)
The following discussion is for the benefit criterion only.
For case 1, obviously, there is no gain to the DM. Since CiHj < RLj , Li j = 0, i ∈ N, j ∈ M (18)
Gi j = 0, i ∈ N, j ∈ M (10) respectively.
L. Wang et al. / Expert Systems With Applications 42 (2015) 9379–9388 9383


Table 2 the jth attribute, satisfying m j=1 w j = 1, j ∈ M. By using the simple
Gain and loss for all possible cases (benefit criteria).
additive weighting method, the overall prospect value of each alter-
Cases Loss Li j Gain Gi j native can be obtained, that is,

Case 1 CiHj < RLj 0.5( CiLj + CiHj )− RLj 0 


m
Oi = v̄i j w j , i ∈ N (24)
Case 2 RHj < CiLj 0 0.5(CiLj + CiHj ) − RHj
j=1
Case 3 CiLj < RLj ≤ CiHj < RHj 0.5(CiLj − RLj ) 0
Obviously, the bigger Oi is, the better alternative Si will be. Therefore,
Case 4 RLj < CiLj ≤ RHj < CiHj 0 0.5(CiHj − RHj )
in accordance with a descending order of the overall prospect values
Case 5 CiLj < RLj < RHj < CiHj 0.5(CiLj − RLj ) 0.5(CiHj − RHj ) of all alternatives, we can determine the ranking order of all alter-
Case 6 RLj ≤ CiLj < CiHj ≤ RHj 0 0 natives and select the desirable alternative from the alternative set
S = {S1 , S2 , . . . , Sn }. The best alternative is the one with the biggest
Table 3 overall prospect value.
Gain and loss for all possible cases (cost criteria).
3.4. TOPSIS method with interval reference point
Cases Gain Gi j Loss Li j

Case 1 CiHj < RLj RLj − 0.5(CiLj + CiHj ) 0 In an emergency, the TOPSIS method with interval reference point
Case 2 RHj < CiLj 0 RHj − 0.5(CiLj + CiHj ) considering the DM’s psychological behavior proposed in this study
Case 3 CiLj < RLj ≤ CiHj < RHj 0.5(RLj − CiLj ) 0
is different from other TOPSIS methods. It is not only suitable to cope
with information inadequacy, uncertainty, and dynamic situations,
Case 4 RLj < CiLj ≤ RHj < CiHj 0 0.5(RHj − CiHj )
but it also takes the DM’s psychological behavior into account. The
Case 5 CiLj < RLj < RHj < CiHj 0.5( RLj − CiLj ) 0.5(RHj − CiHj )
procedure for solving the EDM problem with the proposed TOPSIS
Case 6 RLj ≤ CiLj < CiHj ≤ RHj 0 0 method is similar to the TOPSIS method described in Section 2.2 ex-
cept for Step 1, as follows.
Step 1. According to the positional relationship between the ref-
For case 5, CiLj < RLj < RHj < CiHj , similar to the analysis of cases 3 erence point R j and the interval attribute value Ci j shown in Table 1,
and 4, the gain and loss to the DM are given by Eqs. (19)–(20), respec- calculate Gi j or Li j using Eqs. (10)–(22), and then, construct the deci-
tively, that is, sion matrix GLM.
Step 2. With respect to GLM, using the TOPSIS method described
Gi j = 0.5(CiHj − RHj ), i ∈ N, j ∈ M (19)
in Section 2.2, we may easily obtain the results and ranking order
Li j = 0.5(CiLj − RLj ), i ∈ N, j ∈ M (20) of alternatives, which are the gist of decision making for emergency
response action.
When the possible positional relationship between R j and Ci j is sim-
ilar to case 5, the final outcome is gain or loss, which depends on the 4. Emergency response
summation of Eqs. (19) and (20).
For case 6, since RLj ≤ CiLj < CiHj ≤ RHj , that is, [CiLj , CiHj ] ⊆ [RLj , RHj ], any 4.1. First stage of emergency response
possible value in interval [CiLj , CiHj ] is equally acceptable to the DM
(Bordley & Kirkwood, 2004). Thus, the gain and loss are given by Let t = {t1 , t2 , . . . , tn } be a set of time points at which the DM re-
ceives an alarm or alters the emergency alternative. When an emer-
Gi j = 0, i ∈ N, j ∈ M (21) gency event occurs, the DM receives an alarm at the first time, t1 , but
Li j = 0, i ∈ N, j ∈ M (22) because of information inadequacy and uncertainty about the acci-
dent site, the DM has insufficiently accurate information to respond
respectively. reasonably. At t1 , based on the DM’s own experience, knowledge, and
By Eqs. (10)–(22), the calculation formulae of gain and loss for the descriptions of the scene of the accident described by the people
all possible cases are summarized in Table 2. In general, the crite- who trigger the alarm, the DM estimates the probable losses caused
ria can be classified into two types: benefit and cost (Ma et al., 1999). by the emergency and, at the same time, forms a psychological ref-
Table 2 shows the gain and loss for all possible cases for benefit cri- erence point with respect to probable losses. Then, based on the ref-
teria/attributes. Table 3 shows the gain and loss for cost criteria. Note erence point, the DM activates an alternative to undertake an imme-
that for cost criteria, if Ci j < R j , the DM feels gains, and if Ci j > R j , the diate response. After executing the alternative for a period, there are
DM feels losses. Furthermore, the GLM can be constructed. two types of response results. One is the ideal result, which means
that the emergency alternative has controlled the situation effec-
3.3. Calculation of prospect values and ranking of alternatives tively and losses caused by the emergency event are all under the
DM’s psychological reference point; in other words, the alternative is
Let GLM = (xi j )n×m , where xi j denotes Gi j or Li j . Based on Eq. (1), effective, the losses caused by the emergency event are acceptable to
it is easy to obtain the prospect value of each attribute with respect the DM, and in this situation, the alternative needs no adjustment.
to each alternative, VM = (vi j )n×m , where vi j denotes the jth attribute The second type is a negative-ideal result, which implies that the al-
prospect value with respect to the ith alternative. Since gains or losses ternative has not controlled the escalation and deterioration of the
with respect to different attributes are generally incommensurate, situation effectively and the losses are above the DM’s psychological
they need to be normalized to be transformed into comparable val- reference point, which is not acceptable to the DM. In this situation,
ues. This is achieved by normalizing every element in VM into a cor- the DM should adjust the alternative dynamically in order to avoid
responding element in matrix VM = (v̄i j )n×m by using the following more damages and losses caused by the emergency event and ensure
formula the safety of human life and society.
vi j
v̄i j = , i ∈ N, j ∈ M (23)
v∗j 4.2. Adjustment of the alternative

where v∗j = max{|vi j |}. Let W = (w1 , . . . , wm ) be the relative weight With respect to the negative-ideal result, the DM should ad-
i∈N
vector about the attribute, where w j denotes the relative weight of just the alternative immediately. After executing for a period, at
9384 L. Wang et al. / Expert Systems With Applications 42 (2015) 9379–9388

t2 (t1 < t2 ), the DM has more adequate and accurate information Table 4
Decision data about five alternatives with respect to different crite-
about the emergency event. With respect to the losses caused by the
ria/alternatives.
continuous development of emergencies, the DM dynamically ad-
justs the psychological reference point, and forms a new reference Alternatives Criteria/Attributes
point R˜ j . Based on the real-time losses/costs C˜ j and the new reference Ci1 Ci2 Ci3 Ci4
points R˜ j , the DM adjusts the corresponding emergency alternatives
S1 [30,50] [3,5] [50,100] [40,50]
dynamically, so as to prevent the further deterioration of the situa- S2 [60,80] [6,13] [10 0,20 0] [50,60]
tion timeously and effectively. C˜ j denotes the real-time losses/costs S3 [90,120] [14,20] [20 0,40 0] [60,70]
of the jth attribute value, j ∈ M, and R˜ j denotes the DM’s new refer- S4 [130,160] [21,30] [50 0,10 0 0] [70,80]
S5 [170,200] [31,50] [10 0 0,30 0 0] [80,90]
ence point with respect to the jth attribute, j ∈ M.
In summary, the procedures for adjustment of the alternative are
given as follows.
Step 1. According to the real-time losses/costs C˜ j , the DM forms alternative’s effective control scope is [50,100] and [40,50], re-
new reference points R˜ j , determines the positional relationship be- spectively, that is, C13 = [50, 100] and C14 = [40, 50].
tween Ci j and R˜ j , based on Table 1, calculates Gi j or Li j using S2 : The cost of S2 is [60,80], that is, C21 = [60, 80]. The effec-
Eqs. (10)–(22), and then, constructs the GLM. tive control scope of casualties, the property losses, and the
Step 2. Using Eq. (1), the DM constructs the value matrix VM. negative effects on the environment are [6,13], [10 0,20 0], and
Step 3. The DM constructs normalized value matrix VM = (v̄i j )n×m [50,60], respectively, that is, C22 = [6, 13], C23 = [10 0, 20 0], and
using Eq. (23). C24 = [50, 60].
Step 4. The DM calculates the overall prospect value of each alter- S3 : The cost of S3 is [90,120], that is, C31 = [90, 120]. The effective
native using Eq. (24). control scope of casualties, the property losses, and the neg-
Step 5. The DM determines the ranking order of alternatives ac- ative effects on the environment are [14,20], [20 0,40 0], and
cording to the obtained overall prospect values, and selects the [60,70], respectively, that is, C32 = [14, 20], C33 = [20 0, 40 0],
desirable alternative to deal with the emergency dynamically and and C34 = [60, 70], respectively.
effectively. S4 : The cost of S4 is [130,160], that is, C41 = [130, 160]. The
Note that for convenience and simplification of the problem, this effective control scope of casualties, the property losses,
study considers the adjustment of the alternative only once. If nec- and the negative effects on the environment are [21,30],
essary, the procedures can be repeated many times until the EDM [50 0,10 0 0], and [70,80], respectively, that is, C42 = [21, 30],
problem is solved. C43 = [50 0, 10 0 0], and C44 = [70, 80], respectively.
S5 : The cost of S5 is [170,200], that is, C51 = [170, 200]. The
5. Case studies effective control scope of casualties, the property losses,
and the negative effects on the environment are [31,50],
To illustrate the validity and feasibility of the proposed methods, [10 0 0,30 0 0], and [80,90], respectively, that is, C52 = [31, 50],
this section presents two real cases. The first is a petrochemical plant C53 = [10 0 0, 30 0 0], and C54 = [80, 90], respectively.
fire, adapted from a real fire that occurred in China’s most famous
petrochemical plant. The second is based on a barrier lake emergency The DM provides the relative weight vector of criteria/attributes,
caused by a huge earthquake that occurred in southwestern China. that is, W = (0.2, 0.35, 0.2, 0.25).
The decision data about the five alternatives with respect to dif-
5.1. Case 1: Petrochemical plant fire ferent criteria/attributes are presented in Table 4.

There are many different types of fire based on different ablaze 5.1.2. Emergency response with proposed prospect theory-based
materials. Different measures should be taken for different situations. method
For the features of inflammability, explosibility, diffusivity, and chain 5.1.2.1. First stage of emergency response. The petrochemical plant
reaction, the materials of petrochemical plants are more dangerous broke the rules of safe operating, which resulted in an oil pipe ex-
than those of common factories. Thus, for petrochemical plants, when plosion and, subsequently, a fire. With regard to the description for
a fire breaks out, the DM must obey the principles of immediate re- the first stage of the emergency response in Section 4.1, at t1 , the
sponse, timely rescue, cooling and extinguishment, so as to control DM receives the fire alarm, estimates the situation, and forms a psy-
the situation effectively and prevent it from further escalation and chological reference point with respect to each criterion/attribute,
deterioration. as follows, R1 = [80, 100], R2 = [10, 15], R3 = [30 0, 40 0], and R4 =
[65, 75]. Based on the reference points, the DM activates the alter-
5.1.1. Description of the problem native S3 to respond to the emergency immediately. However, af-
The petrochemical plant broke rules for safe operating, which re- ter executing for a period, at t2 (t1 < t2 ), the cost of the alternative
sulted in an oil pipe explosion and fire. For the petrochemical plant has reached [120,130], that is, C˜1 = [120, 130], the number of casual-
features, the following criteria/attributes are discussed. ties has reached [20,25], that is, C˜2 = [20, 25], the property losses are
more than [50 0,60 0], that is, C˜3 = [50 0, 60 0], and the negative effect
Ci1 : The cost of each alternative (in 10,0 0 0 RMB).
on the environment has reached [70,80], that is, C˜4 = [70, 80].
Ci2 : The number of casualties.
From the response results, the real-time losses are over the ref-
Ci3 : Property loss (in 10,0 0 0 RMB).
erence point, meaning that the alternative S3 has not controlled the
Ci4 : The negative effect on the environment, the values concerning
situation effectively. Thus, in order to control the situation timeously
criterion Ci4 are evaluated by experts on a scale of 0–100 (0: no
and to prevent further escalation and deterioration, adjustment of the
negative effect; 100: serious negative effect)
alternative is necessary.
The emergency alternatives are as follows.
S1 : The cost of S1 is [30,50], that is, C11 = [30, 50]. The alterna-
tive S1 can control the number of casualties effectively in the 5.1.2.2. Adjustment of the alternative. This subsection adjusts the al-
scope of [3,5], that is, C12 = [3, 5]; similarly, with regard prop- ternative to the first-stage emergency response so as to control the
erty losses and the negative effects on the environment, the situation timeously and effectively to avoid further deterioration.
L. Wang et al. / Expert Systems With Applications 42 (2015) 9379–9388 9385

After executing for a period, at t2 (t1 < t2 ), the DM has adequate Table 5
The distance from the ideal and negative-ideal alternatives.
and accurate information about the situation. According to the real-
time losses, the DM forms new reference points as follows: R˜1 = Alternatives
[150, 180], R˜2 = [30, 35], R˜3 = [10 0 0, 150 0], and R˜4 = [40, 50]. Using S1 S2 S3 S4 S5
the method proposed in Section 4.2, the procedures of the adjust-
ment alternative are presented below. D∗i 0.4565 0.3899 0.2955 0.1782 0.0 0 0 0
D− 0.0 0 0 0 0.0730 0.1762 0.2947 0.4565
First, the DM determines the positional relationship between the i

Ci j and the new reference points R˜ j , based on Tables 2 and 3. The GLM
is constructed as follows, Table 6
Relative closeness to the ideal alternative.
⎡ 110 −26 −925 0

Alternatives
⎢ 80 −20.50 −850 5⎥
GLM = ⎢ 45 15⎥
S1 S2 S3 S4 S5
−13 −700
⎣ ⎦
10 −4.50 −250 25 RCi 0.0 0 0 0 0.1577 0.3735 0.6232 1.0 0 0 0
−10 7.50 750 35 Rank 5 4 3 2 1

By using Eq. (1), the value matrix VM is constructed as follows,


The weighted normalized decision matrix V is constructed using
⎡ 65.590 −45.077 −1205.13 0

Eq. (3) as follows,
⎢ 49.403 −36.224 −1114.93 4.189 ⎥ ⎡ 0.153 −0.248 −0.113 0

VM = ⎢ 29.605 −23.824 −932.551 11.136⎥
⎣ ⎦ ⎢ 0.111 −0.196 −0.104 0.027⎥
7.762 −8.977 −361.649 17.546
V = ⎢ 0.063 −0.124 −0.085 0.082⎥
−18.715 6.009 362.08 23.671 ⎣ ⎦
0.014 −0.043 −0.030 0.136
−0.014 0.072 0.091 0.191
According to the VM, the normalized value matrix VM is con-
structed using Eq. (23) as follows, Accordingly, the ideal and negative-ideal solutions can be de-
fined as S∗ = {−0.014, 0.072, 0.091, 0.191} and S− = {0.153, −0.248,
⎡ 1.0 0 0 −1.0 0 0 −1.0 0 0 0
⎤ −0.113, 0}.
⎢ 0.753 −0.804 −0.925 0.177⎥ Euclidean distances from the ideal and negative-ideal alternatives
VM = ⎢ 0.451 −0.529 −0.774 0.470⎥ are calculated using Eqs. (6) and (7), respectively. The results are de-
⎣ ⎦
0.118 −0.199 −0.300 0.741 picted in Table 5.
−0.285 0.133 0.300 1.0 0 0 By applying Eq. (8), the relative closeness to the ideal solution ob-
tained and the results are shown in Table 6.
Furthermore, using Eq. (24), the overall prospect value of each The relative closeness values to the ideal alternative are 0.0 0 0 0 for
alternative is obtained, that is, O1 = −0.3500, O2 = −0.2714, O3 = S1 , 0.1577 for S2 , 0.3735 for S3 , 0.6232 for S4 , and 1.0 0 0 0 for S5 . That is,
−0.1319, O4 = 0.0793, and O5 = 0.2997. Therefore, the ranking order alternative S5 has the greatest value; therefore, it should be the best
of the five alternatives is S5  S4  S3  S2  S1 . Obviously, S5 is the alternative for the emergency response.
most desirable alternative.
5.2. Case 2: barrier lake emergency

To illustrate the validity and feasibility of the proposed methods


5.1.3. TOPSIS method with interval reference point
further, this subsection presents another example based on the back-
According to the description of the problem in Section 5.1, the
ground of a barrier lake emergency caused by a huge earthquake that
same example is used to illustrate the proposed TOPSIS method with
occurred in southwestern China.
interval reference point. The procedures and results are presented be-
low.
5.2.1. Description of the barrier lake problem
First, with respect to the emergency, the DM forms psycholog-
A barrier lake, formed by fallen rocks and a landslide after a huge
ical reference points. According to Table 1, he or she determines
earthquake, threatened the lives and properties of thousands of peo-
the relationship between the interval values Ci j and the reference
ple both upstream and downstream. The DM organized an emergency
points R j . For convenience of computation, the reference points are
expert panel to deal with intractable problems. Based on the analysis
R1 = [150, 180], R2 = [30, 35], R3 = [10 0 0, 150 0], and R4 = [40, 50],
of this panel, four potential emergency states of the barrier lake in
respectively. Based on Tables 2 and 3, the decision matrix GLM is con-
the 72 h after the earthquake emerged, namely,
structed as follows,
⎡ 110 −26 −925 0
⎤ A1 : the dam body of the barrier lake will not break;
A2 : 1/3 of the dam body of the barrier lake will break;
⎢ 80 −20.50 −850 5⎥ A3 : 1/2 of the dam body of the barrier lake will break; and
GLM = ⎢ 45 −13 −700 15⎥
⎣ ⎦ A4 : the entire dam body of the barrier lake will break.
10 −4.50 −250 25
−10 7.50 750 35 To protect lives and properties, there are four emergency alterna-
tives for emergency response as follows,
By using Eq. (2), the normalized matrix is obtained as follows,
S1 : Evacuate people from the most dangerous upstream and
⎡ 0.764 −0.710 −0.564 0
⎤ downstream areas of the barrier lake to safe areas, and inform
⎢ 0.556 −0.560 −0.518 0.109⎥ people in potentially dangerous areas to prepare for evacua-
GLM = ⎢ 0.313 −0.355 −0.427 0.327⎥ tion. At the same time, combine repeated small batch quan-
⎣ ⎦
0.069 −0.123 −0.152 0.546 tities of artificial blasting and excavation of drain grooves to
−0.069 0.205 0.457 0.764 meet the requirements of the discharged barrier lake floods.
9386 L. Wang et al. / Expert Systems With Applications 42 (2015) 9379–9388

Table 7 According to the real-time losses C˜ j , the new reference points R˜ j


Effective control scopes and cost of alternative Si for different
attributes. determine the relationship between Ci j and R˜ j , based on Table 1. Us-
ing Eqs. (10)–(22), the GLM is constructed as follows,
Alternatives Criteria/Attributes ⎡ ⎤
−1250 −500 475
Ci1 Ci2 Ci3
⎢ −750 −250 400⎥
GLM = ⎣
S1 [30 0 0,350 0] [250 0,350 0] [300,350] −250 250 300⎦
S2 [350 0,40 0 0] [350 0,450 0] [350,450] 50 500 200
S3 [40 0 0,450 0] [450 0,550 0] [450,550]
The value matrix VM is constructed using Eq. (1) as follows
S4 [50 0 0,550 0] [550 0,650 0] [550,650]
⎡ ⎤
−1589.790 −684.282 241.133
⎢ −993.662 −361.649 206.935⎥
VM = ⎣
The cost Ci3 of S1 is [300,350] (10,000 RMB), and the effective −361.649 136.197 160.191⎦
control scope of alternative S1 with respect to the number of 32.515 252.396 111.665
people affected Ci1 and the property losses Ci2 are [30 0 0,350 0],
According to the VM, the normalized value matrix VM is constructed
and [250 0,350 0], respectively;
using Eq. (23) as follows,
S2 : Based on S1 , increase the joint scheduling of the reservoir and ⎡ ⎤
hydropower station in the upstream and downstream areas to −1.0 0 0 −1.0 0 0 1.0 0 0
reduce the pressure of the barrier lake. The cost Ci3 of S2 is ⎢−0.625 −0.529 0.858⎥
VM = ⎣
[350,450] (10,0 0 0 RMB), and the effective control scope of al- −0.227 0.199 0.664⎦
ternative S2 with respect to the number of people affected Ci1 0.020 0.369 0.463
and the property losses Ci2 are [350 0, 40 0 0], and [350 0,450 0], Furthermore, using Eq. (24), the overall prospect value of each
respectively; alternative is obtained, that is, O1 = −0.50 0 0, O2 = −0.2204, O3 =
S3 : Based on evacuating people, mobilize large, heavy machinery 0.1448, and O4 = 0.2530. Therefore, the ranking order of the four al-
and implement large-scale blasting to reduce the water level ternatives is S4  S3  S2  S1 . Obviously, S4 is the most desirable al-
of the barrier lake as much as possible to lower the risk of dam ternative.
break. The cost Ci3 of S3 is [450,550] (10,0 0 0 RMB), and the ef-
fective control scope of alternative S3 with respect to the num- 5.2.3. TOPSIS method with interval reference point
ber of people affected Ci1 and the property losses Ci2 are [40 0 0, To illustrate the proposed TOPSIS method with interval refer-
4500], and [4500, 5500], respectively; ence point further, an emergency response of the barrier lake is pro-
S4 : Based on evacuating people, increase the joint scheduling of posed to illustrate the feasibility and validity of the proposed TOPSIS
the reservoir and hydropower station in the upstream and method in this subsection.
downstream areas. Meanwhile, mobilize large, heavy machin- For convenience of computation, the reference points are R1 =
ery and implement large-scale blasting to reduce the water [450 0, 50 0 0], R2 = [40 0 0, 50 0 0], and R3 = [80 0, 10 0 0], respectively.
level of the barrier lake as much as possible to lower the risk The decision matrix GLM is constructed using Eqs. (10)–(22), as fol-
of dam break. The cost Ci3 of S4 is [550,650] (10,0 0 0 RMB), and lows
⎡ ⎤
the effective control scope of alternative S4 with respect to the −1250 −500 475
number of people affected Ci1 and the property losses Ci2 are ⎢ −750 −250 400⎥
GLM = ⎣
[50 0 0, 550 0] and [550 0, 650 0], respectively. −250 250 300⎦
50 500 200
The attribute weights are provided by DM directly, that is, W =
(0.4, 0.35, 0.25). The effective control scopes with respect to different Using Eq. (2), the normalized matrix is obtained as follows
⎡ ⎤
criteria/attributes of each alternative are shown in Table 7. −0.845 −0.632 0.661
⎢−0.507 −0.316 0.557⎥
GLM = ⎣
5.2.2. Emergency response with proposed PT-based method −0.169 0.316 0.418⎦
5.2.2.1. Emergency response at t1 . When the disaster occurs at t1 , the 0.034 0.632 0.279
DM receives an emergency alarm. Because of inadequate information The weighted normalized decision matrix V is constructed using
about the barrier lake, the DM estimates the probable damages and Eq. (3) as follows
losses caused by the emergency, and forms corresponding reference ⎡ ⎤
points with respect to each attribute, that is, R1 = [330 0, 360 0], R2 = −0.338 −0.221 0.165
[280 0, 330 0], and R3 = [230, 30 0]. Based on the reference points, the ⎢−0.203 −0.111 0.139⎥
V=⎣
DM activates the alternative S2 for the emergency. After executing for −0.068 0.111 0.104⎦
a period at t2 ; however, owing the aftershocks and rain, the danger 0.014 0.221 0.070
associated with the barrier lake increases, while the real-time losses Accordingly, the ideal and negative-ideal solutions can be defined as
of the number of people affected C˜1 , the property losses C˜2 , and the S∗ = {0.014, 0.221, 0.070} and S− = {−0.338, −0.221, 0.165}.
cost of the alternative C˜3 reach C˜1 = [380 0, 40 0 0], C˜2 = [350 0, 380 0], Euclidean distances from the ideal and negative-ideal alternatives
and C˜3 = [350, 400], respectively. According to the real-time losses, are calculated using Eqs. (6) and (7), respectively. The results are de-
clearly, the alternative S2 does not control the emergency effectively, picted in Table 8.
and thus, the alternative has to be adjusted so as to prevent the situ- Applying Eq. (8), the relative closeness to the ideal solution ob-
ation escalating and deteriorating further. tained and the results are shown in Table 9.
The relative closeness values to the ideal alternative are 0.0 0 0 0 for
5.2.2.2. Adjustment of the alternative. After executing for a period at S1 , 0.3051 for S2 , 0.7534 for S3 , and 1.0 0 0 0 for S4 . That is, alternative S4
t2 , the DM has adequate and accurate information about the barrier has the greatest value; therefore, it should be the best alternative for
lake emergency. Based on the real-time losses, the DM forms the new emergency response. Note that the result show there is consistency
reference points, that is, R˜1 = [450 0, 50 0 0], R˜2 = [40 0 0, 50 0 0], and between the best alternative S4 and the actual emergency response
R˜3 = [80 0, 10 0 0]. Using the method in Section 4.2, the procedures of alternative of the barrier lake, which verifies the feasibility and valid-
the adjustment alternative are presented below. ity of the proposed approach.
L. Wang et al. / Expert Systems With Applications 42 (2015) 9379–9388 9387

Table 8 into consideration in this study, which are usually employed by DM


Distance from the ideal and negative-ideal alternatives.
under EDM situation.
Alternatives

S1 S2 S3 S4 6. Conclusions

D∗i 0.5733 0.4023 0.1416 0.0 0 0 0


D− 0.0 0 0 0 0.1766 0.4325 0.5733
The existing EDM methods mainly concern approaches to deal-
i
ing with the inadequate and incomplete information in emergen-
cies or selecting the ideal emergency alternatives. They neglect the
Table 9 importance of DMs’ psychological behavior under emergencies, and
Relative closeness to the ideal alternative.
few studies take the DM’s psychological behavior into account. They
Alternatives do not focus on the dynamic features of emergency events and the
S1 S2 S3 S4 DMs’ limited judgments under risk and uncertainty. Based on such
problems, this study proposes a PT-based interval dynamic reference
RCi 0.0 0 0 0 0.3051 0.7534 1.0 0 0 0 point method for EDM. In addition, because existing TOPSIS methods
Rank 4 3 2 1
are seldom concerned with DMs’ psychological behavior and limited
judgments in emergencies, this study proposes a TOPSIS method with
5.3. Discussion interval reference point.
Compared with the existing methods, the proposed PT-based in-
To illustrate the feasibility and validity of the proposed methods, terval dynamic reference point method produces an exact fuzzy es-
two different real emergencies that occurred in China are presented. timate rather than a crisp number as the reference point under
In the actual EDM, the DM selects S5 and S4 in cases 1 and 2, respec- emergencies. There are significant differences between the proposed
tively. These alternatives successfully dealt with the emergencies in PT-based interval dynamic reference point method and the other ver-
the real world. This indicates the effectiveness of the proposed meth- sions. It takes the dynamic features of emergency events into account,
ods in this study. which is closer to actual EDM and easy for DMs to accept and apply. In
Compared with the final ranking of the alternatives using the pro- addition, it has a simpler and faster computation process than other
posed methods in the two emergency examples, it is noteworthy that versions.
the same rankings of alternatives are obtained. In general, according The proposed TOPSIS method with interval reference point mainly
to different emergencies, emergency alternatives should be divided focuses on DMs’ psychological behavior under EDM, which is seldom
into different levels to deal with the corresponding situation, so as considered in the existing TOPSIS methods. Our TOPSIS method is
to control each situation effectively and avoid wasting resources and closer to the actual EDM and is significantly different to other ver-
workforce. By analysis, the same rankings of alternatives are codeter- sions that neglect DMs’ psychological behavior under EDM. Two real
mined by the reference points and the features of gradually strength- emergency examples have been examined using the proposed PT-
ened alternatives. The feasibility and validity of the proposed meth- based interval dynamic reference point method and the proposed
ods are not only verified by each other but also through the actual TOPSIS method, demonstrating their feasibility and validity. In ad-
EDM. dition, based on the interesting final rankings of the alternatives in
Compared with other methods, the advantages of the proposed the examples, each of the proposed methods verifies the feasibility
methods are as follows. and validity of the other. It is expected that the methods developed
in this study may have more potential applications in the near future.
(1) The PT-based interval dynamic reference point method is
based on “bounded rationality” using PT; it fully considers Acknowledgments
the DM’s bounded rationality under risk and uncertainty com-
pared to other traditional methods that neglect the DM’s psy- This work was partly supported by the National Natural Sci-
chological behavior. Compared with the models based on PT in ence Foundation of China (Project no. 71371053), Humanities and
the literature (Fan et al., 2012; Liu et al., 2014; Wang & Wang, Social Science Foundation of the Ministry of Education (Project
2013), the existing methods neglect the features of dynamic No. 14YJC630056), the Specialized Research Fund for the Doc-
and inadequate/incomplete information of emergency events toral Program (SRFDP) of Higher Education of China under grant
as well as DMs’ limited judgments under risk and uncertainty. No. 20123514110012 and the Natural Science Foundation of Fujian
However, the proposed method in this study takes these issues Province (Project No. 2014J01264)
into account. The method is understood easily, acceptable to
the DM, and closer to the actual EDM situation. Because it uses References
a simple additive weighting method, the proposed method has
a simpler and faster computation process than other versions Abdellaoui, M., Bleichrodt, H., & Paraschiv, C. (2007). Loss aversion under prospect the-
(Fan et al., 2012; Liu et al., 2014; Wang & Wang, 2013). These ory:a parameter-free measurement. Management Science, 53(10), 1659–1674.
Abo-Sinna, M. A., & Amer, A. H. (2005). Extensions of TOPSIS for multi-objective large-
improvements expand the scope of PT application.
scale nonlinear programming problems. Applied Mathematics and Computation,
(2) The existing TOPSIS literature, (Chen, 2012; Dymova, 2013a), 162(1), 243–256.
has made a significant contribution to the TOPSIS method. Bell, D. E. (1982). Regret in decision making under uncertainty. Operations research,
However, these studies are a little different from the proposed 30(5), 961–981.
Bell, D. E. (1985). Disappointment in decision making under uncertainty. Operations
TOPSIS method with interval reference point in this study, Research, 33(1), 1–27.
which fully considers the DM’s psychological behavior and his Bleichrodt, H., Schmidt, U., & Zank, H. (2009). Additive utility in prospect theory. Man-
or her limited judgments under risk and uncertainty. The pro- agement Science, 55(5), 863–873.
Booij, A. S., & Van de Kuilen, G. (2009). A parameter-free analysis of the utility of money
posed TOPSIS method in this study is closer to the actual EDM, for the general population under prospect theory. Journal of Economic Psychology,
which is understood easily and applied by and acceptable to 30(4), 651–666.
the DM. Both methods expand the scope of TOPSIS application. Bordley, R. F., & Kirkwood, C. W. (2004). Multiattribute preference analysis with perfor-
mance targets. Operations Research, 52(6), 832–835.
Chen, S. M., & Lee, L. W. (2010). Fuzzy multiple attributes group decision-making
There are some limitations of the proposed methods, for example, based on the interval type-2 TOPSIS method. Expert systems with applications, 37(4),
fuzzy linguistic variables and triangular fuzzy numbers do not take 2790–2798.
9388 L. Wang et al. / Expert Systems With Applications 42 (2015) 9379–9388

Chen, T. Y. (2011). Interval-valued fuzzy TOPSIS method with leniency reduction and Loomes, G., & Sugden, R. (1986). Disappointment and dynamic consistency in choice
an experimental analysis. Applied Soft Computing, 11(8), 4591–4606. under uncertainty. The Review of Economic Studies, 53(2), 271–282.
Chen, T. Y., & Tsao, C. Y. (2008). The interval-valued fuzzy TOPSIS method and experi- Ma, J., Fan, Z. P., & Huang, L. H. (1999). A subjective and objective integrated approach
mental analysis. Fuzzy Sets and Systems, 159(11), 1410–1428. to determine attribute weights. European Journal of Operational Research, 112(2),
Chen, T. Y. (2012). Comparative analysis of SAW and TOPSIS based on interval-valued 397–404.
fuzzy sets: discussions on score functions and weight constraints. Expert Systems Mahdevari, S., Shahriar, K., & Esfahanipour, A. (2014). Human health and safety risks
with Applications, 39(2), 1848–1861. management in underground coal mines using fuzzy TOPSIS. Science of the Total
Chen, T. Y. (2015). The inclusion-based TOPSIS method with interval-valued intuition- Environment, 488, 85–99.
istic fuzzy sets for multiple criteria group decision making. Applied Soft Computing, Mendonça, D., Rush, R., & Wallace, W. A. (20 0 0). Timely knowledge elicitation from ge-
26, 57–73. ographically separate, mobile experts during emergency response. Safety Science,
Cheng, S., Chan, C. W., & Huang, G. H. (2003). An integrated multi-criteria decision 35(1), 193–208.
analysis and inexact mixed integer linear programming approach for solid waste Millar, H., Joseph, A., & Mobolurin, A. (2002). Using formal MS/OR modeling to support
management. Engineering Applications of Artificial Intelligence, 16(5), 543–554. disaster recovery planning. European Journal of Operational Research, 141(3), 679–
Chu, T. C. (2002a). Facility location selection using fuzzy TOPSIS under group decisions. 688.
International journal of uncertainty, fuzziness and knowledge-based systems, 10(06), Mokhtarian, M. N., Sadi-Nezhad, S., & Makui, A. (2014). A new flexible and reliable IVF-
687–701. TOPSIS method based on uncertainty risk reduction in decision making process.
Chu, T. C. (2002b). Selecting plant location via a fuzzy TOPSIS approach. The Interna- Applied Soft Computing, 23, 509–520.
tional Journal of Advanced Manufacturing Technology, 20(11), 859–864. Mo, S., Duan, H., Shen, B., & Wang, D. (2015). Interval Two-stage stochastic integer
Chu, T. C., & Lin, Y. C. (2003). A fuzzy TOPSIS method for robot selection. The Interna- programming for urban water resource management under uncertainty. Journal
tional Journal of Advanced Manufacturing Technology, 21(4), 284–290. of Coastal Research, 73(sp1), 160–165.
Chu, T. C., & Lin, Y. C. (2009). An interval arithmetic based fuzzy TOPSIS model. Expert Nwogugu, M. (2006). A further critique of cumulative prospect theory and related ap-
Systems with Applications, 36(8), 10870–10876. proaches. Applied mathematics and computation, 179(2), 451–465.
De Maio, C., Fenza, G., & Gaeta, M. (2011). A knowledge-based framework for emer- Rainabralimam, B. V., Sreenivasulu, B., & Mallikaijunan, M. M. (1996). Model on-site
gency DSS. Knowledge-Based Systems, 24(8), 1372–1379. emergency plan. Case study: Toxic gas release from an ammonia storage terminal.
Deepa, J., & Sanjay, K. (2015). Interval-valued intuitionistic hesitant fuzzy Cho- Journal of Loss Prevention in the Process Industries, 9(4), 259–265.
quet integral based TOPSIS method for multi-criteria group decision mak- Rashid, T., Beg, I., & Husnine, S. M. (2014). Robot selection by using generalized interval-
ing. European Journal of Operational Research, Available online 2 July 2015, valued fuzzy numbers with TOPSIS. Applied Soft Computing, 21, 462–468.
http://dx.doi.org/10.1016/j.ejor.2015.06.047. Schmidt, U., Starmer, C., & Sugden, R. (2008). Third-generation prospect theory. Journal
Dymova, L., Sevastjanov, P., & Tikhonenko, A. (2013a). An approach to generalization of of Risk and Uncertainty, 36(3), 203–223.
fuzzy TOPSIS method. Information Sciences, 238, 149–162. Schmidt, U., & Zank, H. (2008). Risk aversion in cumulative prospect theory. Manage-
Dymova, L., Sevastjanov, P., & Tikhonenko, A. (2013b). A direct interval extension of ment Science, 54(1), 208–216.
TOPSIS method. Expert Systems with Applications, 40(12), 4841–4847. Schmidt, U., & Zank, H. (2012). A genuine foundation for prospect theory. Journal of Risk
Dymova, L., Sevastjanov, P., & Tikhonenko, A. (2015). An interval type-2 fuzzy extension and Uncertainty, 45(2), 97–113.
of the TOPSIS method using alpha cuts. Knowledge-Based Systems, 83, 116–127. Sun, B., Ma, W., & Zhao, H. (2013). A fuzzy rough set approach to emergency mate-
Ergu, D., Kou, G., & Shi, Y. (2011). Analytic network process in risk assessment and de- rial demand prediction over two universes. Applied Mathematical Modelling, 37(10),
cision analysis. Computers & Operations Research, 42, 58–74. 7062–7070.
Ergu, D., Zhang, M., Guo, Z., & Yang, X. (2014). Consistency simulation and optimization Tsaur, R. C. (2011). Decision risk analysis for an interval TOPSIS method. Applied Math-
for HPIBM model in emergency decision making. Procedia Computer Science, 31, ematics and Computation, 218(8), 4295–4304.
558–566. Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of
Fan, Z. P., Liu, Y., & Shen, R. J. (2012). Risk decision analysis method for emergency choice. Science, 211(4481), 453–458.
response based on prospect theory. Systems Engineering-Theory & Practice, 32(5), Tversky, A., & Kahneman, D. (1991). Loss aversion in riskless choice: a reference-
977–984. dependent model. The Quarterly Journal of Economics, 106(4), 1039–1061.
Fan, Z. P., Zhang, X., & Chen, F. D. (2013). Multiple attribute decision making consider- Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: cumulative repre-
ing aspiration-levels: a method based on prospect theory. Computers & Industrial sentation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297–323.
Engineering, 65(2), 341–350. Wakker, P. P. (2010). Prospect theory: For risk and ambiguity. Cambridge: Cambridge Uni-
Fu, P. P., Wu, C., & Tang, J. (2012). Unconventional emergency management based on versity Press.
intelligent group decision-making methodology. Advances in Information Sciences Wang, L., & Wang, Y. M. (2013). Study on the emergency decision method of dynamic
& Service Sciences, 4(7), 208–216. reference point based on prospect theory. Chinese Journal of Management Science,
Hämäläinen, R. P., Lindstedt, M. R. K., & Sinkko, K. (20 0 0). Multiattribute risk analysis S1, 132–140.
in nuclear emergency management. Risk Analysis, 20(4), 455–468. Wang, Y. M., & Elhag, T. (2006). Fuzzy TOPSIS method based on alpha level sets with
Hwang, C. L., & Yoon, K. (1981). Multiple Attributes Decision Making Methods and Appli- an application to bridge risk assessment. Expert systems with applications, 31(2),
cations. Berlin Heidelberg: Springer. 309–319.
Ju, Y., Wang, A., & You, T. (2015). Emergency alternative evaluation and selection based Wu, G., & Markle, A. B. (2008). An empirical test of gain-loss separability in prospect
on ANP, DEMATEL, and TL-TOPSIS. Natural Hazards, 75(2), 347–379. theory. Management Science, 54(7), 1322–1335.
Jahanshahloo, G. R., Lotfi, F. H., & Davoodi, A. R. (2009). Extension of TOPSIS for Wu, B., & Zhao, L. D. (2010). Knowledge model of emergency decision-making based
decision-making problems with interval data: interval efficiency. Mathematical and on knowledge supply and demand. E-business Technology and Strategy, 113, 305–
Computer Modelling, 49(5), 1137–1142. 317.
Kahneman, D., & Tversky, A. (1979). Prospect theory: an analysis of decision under risk. Xiong, W., & Li, J. L. (2011). An emergency local group decision-making model based on
Econometrica, 47(2), 263–291. variable precision rough set, management and service science. 2011 International
Körte, J. (2003). Risk-based emergency decision support. Reliability Engineering & Sys- Conference on. IEEE, 1–4.
tem Safety, 82(3), 235–246. Xu, X., Liang, D., Chen, X., & Zhou, Y. (2015). A risk elimination coordination method for
Krohling, R. A., & Campanharo, V. C. (2011). Fuzzy TOPSIS for group decision making: a large group decision-making in natural disaster emergencies. Human and Ecologi-
case study for accidents with oil spill in the sea. Expert Systems with Applications, cal Risk Assessment: An International Journal, 21(5), 1314–1325.
38(4), 4190–4197. Yang, J., & Xu, C. (2012). Emergency decision engineering model based on sequential
Lai, Y. J., Liu, T. Y., & Hwang, C. L. (1994). Topsis for MODM. European Journal of Opera- games. Systems Engineering Procedia, 5, 276–282.
tional Research, 76(3), 486–500. Ye, F. (2010). An extended TOPSIS method with interval-valued intuitionistic fuzzy
Levy, J. K., & Taji, K. (2007). Group decision support for hazards planning and emergency numbers for virtual enterprise partner selection. Expert Systems with Applications,
management: a group analytic network process (GANP) approach. Mathematical 37(10), 7050–7055.
and Computer Modelling, 46(7), 906–917. Yu, L., & Lai, K. K. (2011). A distance-based group decision-making methodology for
Liu, Y., Fan, Z. P., & Yuan, Y. (2014). A FTA-based method for risk decision-making in multi-person multi-criteria emergency decision support. Decision Support Systems,
emergency response. Computers & Operations Research, 42, 49–57. 51(2), 307–315.
Liu, Y., Fan, Z. P., & Zhang, Y. (2014). Risk decision analysis in emergency response: a Yue, Z. (2014). TOPSIS-based group decision-making methodology in intuitionistic
method based on cumulative prospect theory. Computers & Operations Research, fuzzy setting. Information Sciences, 277, 141–153.
42, 75–82. Zhao, J., Jin, T., & Shen, H. (2007). A case-based evolutionary group decision support
Liu, Y., Fan, Z. P., & Yuan, Y. (2013). Multiple attributes risk decision making method for method for emergency response. Intelligence and Security Informatics (pp. 94–104).
emergency response. Operations Research and Management Science, 01, 23–28. Berlin Heidelberg: Springer.
Loomes, G., & Sugden, R. (1982). Regret theory: an alternative theory of rational choice Zhou, C., Wang, H., & Zhuo, H. H. (2014). A multi-agent coordinated planning approach
under uncertainty. The Economic Journal, 92, 805–824. for deadline required emergency response tasks. IET Control Theory & Applications,
9(3), 447–455.

You might also like