Вы находитесь на странице: 1из 16

Information Sciences 324 (2015) 4459

Contents lists available at ScienceDirect

Information Sciences
journal homepage: www.elsevier.com/locate/ins

On multicriteria decision making under conditions of


uncertainty
J.G. Pereira Jr.a, P.Ya. Ekelb,c,, R.M. Palharesd, R.O. Parreirasb,c
a
Graduate Program in Electrical Engineering, Federal University of Minas Gerais, Av. Antnio Carlos, 6627, 31270-901 Belo Horizonte, MG,
Brazil
b
Graduate Program in Electrical Engineering, Pontical Catholic University of Minas Gerais, Av. Dom Jos Gaspar, 500, 30535-630 Belo
Horizonte, MG, Brazil
c
ASOTECH, Av. Prudente de Morais, 290/507, 30380-002 Belo Horizonte, MG, Brazil
d
Department of Electronics Engineering, Federal University of Minas Gerais, Av. Antnio Carlos, 6627, 31270-901 Belo Horizonte, MG, Brazil

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

Article history: This paper deals with multicriteria decision making problems under conditions of uncer-
Received 28 October 2014 tainty. The main its contribution is the consideration of choice criteria of the classic approach
Revised 23 May 2015
to handle information uncertainty in monocriteria decision making as objective functions
Accepted 11 June 2015
within the framework of multiobjective models, whose analysis generates harmonious so-
Available online 2 July 2015
lutions. Such consideration of choice criteria is of a fundamental character and allows one to
Keywords: modify the generalization, originally proposed by Ekel, Martini, and Palhares (2008), of the
Multicriteria decision making classic approach to handle information uncertainty for solving multicriteria problems. The
Information uncertainty modication permits one to overcome limitations of the indicated generalization, which can
Payoff matrix lead to contradictory decisions. Details of using the modication in a general scheme of mul-
Choice criterion ticriteria decision making under uncertainty are presented. The general scheme is focused on
Fuzzy decision making the use of available quantitative information to the highest degree to reduce decision uncer-
tainty regions. However, if the problem solving capacity concerning quantitative information
processing does not allow one to obtain unique solutions, the general scheme assumes the use
of qualitative information based on knowledge, experience, and intuition of experts involved
in the decision making process. Examples are presented to illustrate the modication of gen-
eralizing the classic approach as well as the general scheme of multicriteria decision making
under information uncertainty.
2015 Elsevier Inc. All rights reserved.

1. Introduction

In the process of posing and solving a wide range of problems of an optimization character related to the design, planning,
operation, and control of complex systems one inevitably encounters diverse types of uncertainty [1,11,19,38,45]. The authors of
[8,38] distinguish two classes of uncertainties in decision making: internal uncertainties (related to decision maker (DM) values
and judgments) and external uncertainties (dened by environmental conditions lying beyond the control of the DM). Although
this paper deals with both kinds of uncertainties, the main its purpose is to handle the external uncertainties in problems of a
multicriteria nature. Taking this into account, we can list two categories of situations, which require the use of a multicriteria
approach [12,33]:


Corresponding author at: Graduate Program in Electrical Engineering, Pontical Catholic University of Minas Gerais, Av. Dom Jos Gaspar, 500, 30535-630
Belo Horizonte, MG, Brazil. Tel.: +55 31 3319 4305; fax: +55 31 3319 4225.
E-mail address: ekel@pucminas.br (P.Ya. Ekel).

http://dx.doi.org/10.1016/j.ins.2015.06.013
0020-0255/ 2015 Elsevier Inc. All rights reserved.
J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459 45

problems in which solution consequences cannot be estimated on the basis of using a single objective. These problems are
related to the analysis of models which include economic as well as physical indices (when alternatives cannot be reduced to
a comparable form) or indices whose cost estimations are hampered (for example, many power engineering problems are to
be considered on the basis of technological, economical, ecological, and social nature criteria);
problems that can be solved on the basis of using a single objective or even several objectives. However if the uncertainty of
information does not permit one to derive unique solutions, then these problems can be rewritten as multicriteria problems
by including additional criteria, which may have a qualitative nature, being based on the knowledge, experience, and intuition
of involved experts (for example, complexity of maintenance, attractiveness of investments, etc.). This can serve as a useful
means to contract the corresponding decision uncertainty regions.

Therefore, two classes of multicriteria models can be constructed [12,33]: X,M, referring to multiobjective models, and
X,R, referring to multiattribute models.
When analyzing X,M models, a vector of objective functions F (X ) = {F1 (X ), . . . ,Fq (X )} is considered and the problem consists
in simultaneous optimizing all objective functions, i.e.,

Fp (X ) extr, p = 1, . . . ,q (1)
XL

where L is a feasible region in Rn .


An important step in analyzing the problem (1) is the determination of a set of Pareto solutions  L [31]. This step is useful.
However, it does not permit one to obtain unique solutions. As consequence, it is necessary to choose a particular Pareto solution
taking into account the information provided by the DM. Three approaches to using this information are classied in [6,17,33]: a
priori, a posteriori, and adaptive.
When analyzing multiobjective problems, it is necessary to develop answers to some specic questions. Among these ques-
tions, it is important to raise the normalization of objective functions, consideration of the importance or priority of each objec-
tive function, and the selection of a principle of optimality. The answers to these questions and, subsequently, the development of
multiobjective methods are carried out in several directions ([9,30,36], for instance). But, without going into further discussions
of that subject, it should be stressed that an important issue in the multiobjective analysis is the quality of obtained solutions.
The quality is considered high if levels of satisfying objective functions are equal or close to each other, when all objectives have
the same importance (refer to the concept of harmonious solutions [12,13]). This concept can be extended to handle objective
functions with different importance levels [33]. At this point, it is important to stress the validity and advisability of the direction
related to the principle of guaranteed result [12,33], which can be implemented by applying the BellmanZadeh approach to de-
cision making in a fuzzy environment [2,34,44]. Its use permits one to realize an effective (from the computational standpoint)
as well as rigorous (from the standpoint of obtaining solutions X 0  L) method of analyzing multiobjective models. Its ap-
plication allows one to preserve a natural measure of uncertainty in decision making and to take into account indices, criteria,
and constraints of qualitative character.
Diverse aspects of utilizing the BellmanZadeh approach in the analysis of X,M models are discussed in [12,17,33]. The use
of the results of [12,17,33] for solving diverse classes of power engineering problems is considered in [4,5,14].
On the other hand, the author of [38] indicates that many multicriteria models are based essentially on deterministic evalu-
ations of the consequences of each action in terms of each criterion, possibly subjecting the nal results and recommendations
to a degree of sensitivity analysis. The use of such an approach may be justied when the primary source of complexity in deci-
sion making is related to the multicriteria nature of the problem rather than to the uncertain nature of individual consequences.
However, in the situations when risks and uncertainties are as critical as issues of conicting goals (for examples, [29,37,38]),
more formal uncertainty modeling is needed [38]. This important consideration is consistent with the view of the authors of
[33], where two principal ways of solving problems under conditions of uncertainty are distinguished. In applying the rst way,
one obtains (at least, theoretically) an exact solution for xed values of the uncertain parameters, and then estimates its stability
for variations of these parameters (for example, by performing multivariant computations). The second way presupposes the
tracking of the effect of the uncertainty at all stages along the path toward the nal decision. This way is more complicated than
the rst one, but is also more fruitful and highly promising. It allows one to increase the adequacy of built models and, as conse-
quence, the credibility and factual eciency of decisions based on their analysis. Taking this into account, it should be noted that
the results of [15,16] may serve as a methodological tool for implementing the second way in the consideration of uncertainty in
multicriteria decision making.
The results of [15,16] are associated with the generalization of the classic approach [3,28,35] to deal with the information
uncertainty, based on building and analyzing payoff matrices reecting effects which can be obtained for different combinations
of solution alternatives and the so-called states of nature or scenarios, in monocriteria decision making to multicriteria problems.
The results of [15,16] are based on combining two branches of mathematics of uncertainty: elements of game theory
and fuzzy set theory and do not t the general approaches to deal with external uncertainties in the multicriteria analy-
sis, classied and analyzed in [8,38] and discussed in more recent works [10,20]. These results, combined with the anal-
ysis of X,M models and X,R models (related to individual or group decision making based on fuzzy preference mod-
eling [18,21,22,24,32,33], for instance; considering that X,R models are used for analyzing multiattribute problems, other
approaches [26,27,42], for example, may be applied as well) served for developing a general scheme of multicriteria decision
46 J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459

making under conditions of uncertainty. This scheme allows one to use available quantitative information in a maximal degree
to reduce the decision uncertainty regions. If the resolving capacity of procedures of formal information processing does not per-
mit one to obtain unique solutions, then the scheme presumes the application of qualitative information based on knowledge,
experience, and intuition of experts.
Examples of applying the generalization of the classic approach dealing with the uncertainty of information to multicriteria
decision making are given in [15,16,33]. However, our experience in its use has revealed some limitations, which can lead to
contradictory decisions, discussed below. The overcoming of these limitations is based on the proposed approach which permits
the simultaneous consideration of choice criteria of the classic approach [3,28,35] as objective functions within the framework
of X,M models, whose analysis provides the harmonious solutions. Such consideration of choice criteria of the classic approach
[3,28,35] is of a fundamental character and serves as the basis for modifying and improving the results of [15,16]. This is convinc-
ingly demonstrated by applications examples.

2. Classic approach to dealing with the uncertainty of information and its generalization to multicriteria decision making
problems

The classic approach [3,28,35] to dealing with the uncertainty of information is based on the assumption that the analysis is
carried out for a given number K of solution alternatives Xk , k = 1, . . . , K and a given number S of representative combinations of
initial data (the states of nature or scenarios) Ys , s = 1, . . . , S, which dene the corresponding payoff matrix (the rst six columns
of Table 1). The payoff matrix reects effects (or consequences) of one or other action Xk , i = 1, . . . , K for the corresponding state
of nature Ys , s = 1, . . . , S.
The analysis of payoff matrices and choice of the rational solution alternatives are based on the use of choice criteria [3,28,35].
The application of the choice criteria of Wald, Laplace, Savage, and Hurwicz, which are of a general character, is discussed in the
present paper. There exist other choice criteria (for example, criteria of Bayes, maximum probability, minimum dispersion, max-
imum measure of Bayesian sets, maximum integral potency, Menges, etc. [23,40]). However, these criteria suppose the certain
information situations on the states of the nature.
To better understand the use of Walds, Laplaces, Savages, and Hurwiczs criteria, Table 1 includes the following characteristic
estimates for the given solution alternative:

the minimum objective function level

F min (Xk ) = min F (Xk , Ys ) (2)


1sS

which is the most optimistic estimate if the objective function is to be minimized or the most pessimistic estimate if the
objective function is to be maximized;
the maximum objective function level

F max (Xk ) = max F (Xk , Ys ) (3)


1sS

which is the most optimistic estimate for the maximized objective function or the most pessimistic estimate if the objective
function is to be minimized;
the average objective function level

1
S
F (Xk ) = F (Xk , Ys ) (4)
S
s=1

the maximum risk (regret) level:

Rmax (Xk ) = max R(Xk , Ys ) (5)


1sS

where R(Xk , Ys ) is an over-expenditure which takes place under combination of the state of nature Ys and the choice of the
solution alternative Xk instead of the solution alternative that is locally optimal for the given Ys .

Table 1
Payoff matrix with characteristic estimates.

Y1  Y2  YS F min (Xk ) F max (Xk ) F (Xk ) Rmax (Xk )

X1 F (X1 ,Y1 )  F (X1 ,Ys )  F (X1 ,YS ) F min (X1 ) F max (X1 ) F (X1 ) Rmax (X1 )
         
Xk F (Xk ,Y1 )  F (Xk ,Ys )  F (Xk ,YS ) F min (Xk ) F max (Xk ) F (Xk ) Rmax (Xk )
         
XK F (XK ,Y1 )  F (XK ,Ys )  F (XK ,YS ) F min (XK ) F max (XK ) F (XK ) Rmax (XK )
F min (Y1 )  F min (Ys )  F min (YS )
J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459 47

To determine the risks R(Xk , Ys ), one needs to dene the minimum value of the objective function (if it is to be minimized, as
in Table 1) for each combination of the state of nature Ys (for each column of the payoff matrix):

F min (Ys ) = min F (Xk , Ys ). (6)


1kK

On the other hand, if the objective function is to be maximized, it is necessary to dene its maximum value for each combination
of the state of nature Ys (for each column of the payoff matrix):

F max (Ys ) = max F (Xk , Ys ). (7)


1kK

The risk for any solution alternative Xk and any state of nature Ys can be evaluated as

R(Xk , Ys ) = F (Xk , Ys ) F min (Ys ) (8)

if the objective function is to be minimized or

R(Xk , Ys ) = F max (Ys ) F (Xk , Ys ) (9)

if it is to be maximized.
The choice criteria, which are based on the use of the characteristic estimates, are represented below under the assumption
that the objective function is to be minimized.
The choice criterion of Wald uses the estimate F max (Xk ) and permits one to choose the solution alternatives X W , for which
the estimate is minimum:

min F max (Xk ) = min max F (Xk , Ys ). (10)


1kK 1kK 1sS

The use of this criterion generates solution alternatives, assuming the most unfavorable combination of initial data. It guar-
antees that the objective function level is not greater than a certain value at any possible future conditions. This is its dignity
[3]. On the other hand, the orientation on the most unfavorable combination of initial data is extremely cautious (pessimistic or
conservative) [3].
The choice criterion of Laplace uses the estimate F (Xk ) and is oriented to choose the solution alternatives X L , for which the
estimate is minimum:

1
S
min F (Xk ) = min F (Xk , Ys ). (11)
1kK 1kK S
s=1

This criterion corresponds to the principle of insucient reason [3], i.e., to the assumption that we have no basis to distin-
guish one or another combination of initial data. Thus it is necessary to act as they are equally probable. This is its drawback.
However, the average score is suciently important.
The choice criterion of Savage is associated with the use of estimate Rmax (Xk ) and allows one to choose the solution alterna-
tives X S , for which the estimate is minimum:

min Rmax (Xk ) = min max R(Xk , Ys ). (12)


1kK 1kK 1sS

As in the case of the choice criterion of Wald, the use of (12) is based on the minimax principle. Therefore, the choice criterion
of Savage can also be considered conservative. However, experience [3] shows that the recommendations based on applying (12)
are mismatched with the decisions obtained with the use of (10). Operating with values of Rmax (Xk ), we obtain a slightly different
evaluation of the situation, which could lead to more daring (less conservative) recommendations.
Finally, the choice criterion of Hurwicz utilizes a convex combination of F max (Xk ) and F min (Xk ) and permits one to choose the
solution alternatives X H , for this combination is minimum:
   
min F max (Xk ) + (1 )F min (Xk ) = min max F (Xk , Ys ) + (1 ) min F (Xk , Ys ) (13)
1kK 1kK 1sS 1sS

where [0, 1] is the index pessimismoptimism whose magnitude is dened by the DM. If = 1, the choice criterion of
Hurwicz is turned into the choice criterion of Wald and if = 0, (13) is turned in the criterion of extreme optimism (minmin)
for which the most favorable combination of initial data is assumed. When 0 < < 1, we obtain something an average and this
is attractiveness of the criterion of Hurwicz. The author of [3] recommends to choose its range from 0.5 to 1.
The choice criteria discussed above have found wide practical applications (for instance, [3,7]) for monocriteria decision
making under conditions of uncertainty. The authors of [25,41,43] have applied fuzzy set theory elements to improve the classic
approach to dealing with information uncertainty, but only in the monocriteria statement.
The choice criteria (10)(13) have also been used in [15,16] if more than one objective function is considered.
In the sequel, we present the information on the use of the BellmanZadeh approach to decision making in a fuzzy envi-
ronment [2,34,44] for solving multiobjective problems and, at the same time, for generalizing the classic approach to deal with
information uncertainty.
48 J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459

When using the BellmanZadeh approach, objective functions Fp (X ), p = 1, . . . , q are replaced by fuzzy sets A p =
{X, A p (X )}, X L, p = 1, . . . , q where A p (X ) is the membership function of A p [34,44]. A fuzzy solution D is dened as
q
D= A
p=1 p
with the membership function

D (X ) = min A p (X ), X L. (14)
1pq

The use of (14) allows one to get the solution

maxD (X ) = max min A p (X ). (15)


XL 1pq

Therefore, the problem (1) is reduced to search for

X 0 = argmax min A p (X ). (16)


XL 1pq

To obtain (16), one needs to build A p (X ), p = 1, . . . , q, which reect a degree of achieving own optima by Fp (X ), X L, p =
1, . . . , q. This condition is satised if one chooses [33]:
 p
maxFp (X ) Fp (X )
A p (X ) = (17)
maxXL Fp (X ) minXL Fp (X )
for minimized objective functions or
 p
Fp (X ) minXL Fp (X )
A p (X ) = (18)
maxXL Fp (X ) minXL Fp (X )
for maximized ones. In (17) and (18), p , p = 1, . . . , q are the importance factors for the corresponding objective functions.
As it was indicated above, the application of the BellmanZadeh approach to decision making in a fuzzy environment provides
constructive lines in obtaining harmonious multicriteria solutions on the basis of analyzing associate maxmin problems (15). Thus
it is possible to propose the generalization of the classic approach to deal with information uncertainty, applying the Bellman
Zadeh approach to decision making in a fuzzy environment. Naturally, if there are q objective functions then q payoff matrices
are constructed and analyzed.
Applying (17) for minimized objective functions or (18) for maximized ones, it is possible to build the modied (normalized)
payoff matrix for the pth objective function presented in Table 2.
The availability of q modied payoff matrices permits one to construct the aggregated payoff matrix presented in Table 3 by
applying (14).
The characteristic estimates of Table 3 are the following: the maximum max D
(Xk ) = max1sS D (Xk , Ys ), minimum
1
S
min
D
(X k ) = min (X
1sS D k s, Y ), and average (X
D k ) = S (X
s=1 D k s , Y ) membership function levels; the maximum risk level,
dened as in (5) with R(Xi , Ys ) = max
D
(Ys ) (X
D i , Ys ) where max (Y ) = max
D s
1kK D k(X , Ys ).
These estimates serve for the choice criteria used under the generalization of the classic approach to deal with information
uncertainty to multicriteria decision making [15,16]. For instance, the modied choice criterion of Laplace has the following form
(see [15,16] for other modied choice criteria):

1
S
max D (Xi ) = max min A p (Xk , Ys ). (19)
1kK 1kK S 1pq
s=1

However, one can point out some limitations of this generalization. For example, one can analyze a hypothetical bicriteria
problem related to minimizing objective functions. The corresponding payoff matrices with characteristic estimates (associated
with the use of the choice criterion of Laplace) are presented in Tables 4 and 5.
The modied payoff matrices for the objective functions are presented in Tables 6 and 7. Finally, the aggregated payoff matrix
with characteristic estimates is given in Table 8.
Notice that the solution X2 is better than X1 when applying the criterion of Laplace (from the point of view of both cri-
teria Tables 4 and 5). However, the analysis of the aggregated payoff matrix (Table 8) shows that the best alternative is X1 .
Although this example is associated with utilizing the choice criterion of Laplace, the use of other choice criteria often leads

Table 2
Modied payoff matrix for the pth objective function.

Y1  Ys  YS

X1 A p (X1 ,Y1 )  A p (X1 ,Ys )  A p (X1 ,YS )


     
Xk A p (Xk ,Y1 )  A p (Xk ,Ys )  A p (Xk ,YS )
     
XK A p (XK ,Y1 )  A p (XK ,Ys )  A p (XK ,YS )
J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459 49

Table 3
Aggregated payoff matrix with characteristic estimates.

Y1  Ys  YS max
D (Xk ) min
D (Xk ) D (Xk ) Rmax (Xk )

X1 D (X1 ,Y1 )  D (X1 ,Ys )  D (X1 ,YS ) max


D (X1 ) min
D (X1 ) D (X1 ) Rmax (X1 )
         
Xk D (Xk ,Y1 )  D (Xk ,Ys )  D (Xk ,YS ) max
D (Xk ) min
D (Xk ) D (Xk ) Rmax (Xk )
         
XK D (XK ,Y1 )  D (XK ,Ys )  D (XK ,YS ) max
D (XK ) min
D (XK ) D (XK ) Rmax (XK )
max
D (Ys ) max
D (Y1 )  max
D (Ys )  max
D (YS )

Table 4
Payoff matrix with characteristic estimates for the rst objective function.

Y1 Y2 F (Xk )

X1 9.00 9.00 9.00


X2 4.20 11.49 7.80
X3 15.00 7.80 11.40
X4 3.00 13.80 8.40

Table 5
Payoff matrix with characteristic estimates for the second objective function.

Y1 Y2 F (Xk )

X1 8.40 14.80 11.60


X2 13.20 5.20 9.20
X3 2.00 18.00 10.00
X4 11.60 13.20 12.40

Table 6
Modied payoff matrix for the rst objective function.

Y1 Y2

X1 0.50 0.50
X2 0.90 0.30
X3 0 0.60
X4 1.00 0.10

Table 7
Modied payoff matrix for the second objective function.

Y1 Y2

X1 0.60 0.20
X2 0.30 0.90
X3 1.00 0
X4 0.40 0.30

Table 8
Aggregated payoff matrix with characteristic estimates.

Y1 Y2 D (Xk )

X1 0.50 0.20 0.35


X2 0.30 0.30 0.30
X3 0 0 0
X4 0.40 0.10 0.25
50 J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459

to similar results. Considering this, the main objective of the present paper is to improve the results of [15,16] to overcome these
contradictions.

3. Choice criteria as objective functions in multicriteria decision making under information uncertainty

The classic approach to deal with the uncertainty of information is associated with analyzing the problems (10)(13) for a
given objective function in an environment with several states of nature Ys , s = 1, . . . , S. Therefore, considering the choice criteria
of Wald, Laplace, Savage, and Hurwicz, respectively, as objective functions, one gets:

F W (Xk ) = F max (Xk ) = max F (Xk , Ys ); (20)


1sS

1
S
F L (Xk ) = F (Xk ) = F (Xk , Ys ); (21)
S
s=1

F S (Xk ) = Rmax (Xk ) = max R(Xk , Ys ); (22)


1sS

F H (Xk ) = F max (Xk ) + (1 )F min (Xk ) = max F (Xk , Ys ) + (1 ) min F (Xk , Ys ). (23)
1sS 1sS

This consideration of the choice criteria of the classic approach permits one to construct q problems, generally, including four
or less objective functions (if not all choice criteria are used in the analysis) as follows:

Fr,p (X ) extr, r = 1, . . . , t 4, p = 1, . . . , q (24)


XL

where the objective functions are: F1,p (X ) = FpW (Xk ), F2,p (X ) = FpL (Xk ), F3,p (X ) = FpS (Xk ), and F4,p (X ) = FpH (Xk ).
Thus, the analysis of the solution alternatives and consequent choice of rational solution alternatives can be realized within
the framework of the X, M models.
Applying (17) to construct the membership functions for Fr,p (X ), r = 1, . . . , t, p = 1, . . . , q, one can solve the problem (15) for the
solution alternatives Xk ,k = 1, . . . , K. The analysis, realized in this way, guarantees the choice of the rational solution alternatives
in accordance with the principle of the Pareto optimality [31] and allows one to overcome the limitations of the generalization of
the classic approach to deal with information uncertainty to multicriteria decision making, as previously discussed. Considering
this, the payoff matrix with the characteristic estimates (Table 1) is presented as the payoff matrix with the choice criteria
estimates for p = 1, . . . , q (Table 9) or for simplicity as the matrix of the choice criteria estimates p = 1, . . . , q in Table 10.
Therefore, using q matrices for the choice criteria estimates, we can construct q modied matrices of the choice criteria
estimates when applying (17), as shown in Table 11.
Finally, with those q modied matrices of the choice criteria estimates, after applying (14), we can construct the aggregated
matrix of the choice criteria estimates, as presented in Table 12. This matrix includes the estimates calculated on the basis of (15)
and used to choose the solution alternatives.
Recalling the example presented in previous section, we can nd the modied matrix of the choice criteria estimates for the
rst objective function (Table 13). At the same time, the modied matrix of the choice criteria estimates for the second objective
function is given in Table 14. The aggregation of these modied matrices of choice criteria estimates is presented in Table 15, and
generates the solution X L = {X2 } based on the use of the choice criterion of Laplace.
Taking into account the results presented above, it is possible to suggest the general scheme of multicriteria decision making
under conditions of information uncertainty which modies the general scheme proposed in [15,16] and, in the general case, is
associated with the following stages:

Table 9
Payoff matrix with choice criteria estimates for the pth objective function.

Y1  Ys  YS FpW (Xk ) FpL (Xk ) FpS (Xk ) FpH (Xk )

X1 Fp (X1 ,Y1 )  Fp (X1 ,Ys )  Fp (X1 ,YS ) FpW (X1 ) FpL (X1 ) FpS (X1 ) FpH (X1 )
         
Xk Fp (Xk ,Y1 )  Fp (Xk ,Ys )  Fp (Xk ,YS ) FpW (Xk ) FpL (Xk ) FpS (Xk ) FpH (Xk )
         
XK Fp (XK ,Y1 )  Fp (XK ,Ys )  Fp (XK ,YS ) FpW (XK ) FpL (XK ) FpS (XK ) FpH (XK )
min1kK FpW (Xk ) min1kK FpL (Xk ) min1kK FpS (Xk ) min1kK FpH (Xk )
max1kK FpW (Xk ) max1kK FpL (Xk ) max1kK FpS (Xk ) max1kK FpH (Xk )
J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459 51

Table 10
Matrix of choice criteria estimates for the pth objective function.

FpW (Xk ) FpL (Xk ) FpS (Xk ) FpH (Xk )

X1 FpW (X1 ) FpL (X1 ) FpS (X1 ) FpH (X1 )


    
Xk FpW (Xk ) FpL (Xk ) FpS (Xk ) FpH (Xk )
    
XK FpW (XK ) FpL (XK ) FpS (XK ) FpH (XK )
min1kK FpW (Xk ) min1kK FpL (Xk ) min1kK FpS (Xk ) min1kK FpH (Xk )
max1kK FpW (Xk ) max1kK FpL (Xk ) max1kK FpS (Xk ) max1kK FpH (Xk )

Table 11
Modied matrix of choice criteria estimates for the pth objective function.

W
A p (Xk ) LA p (Xk ) SA p (Xk ) HAp (Xk )

X1 W
A p (X1 ) LA p (X1 ) SA p (X1 ) HAp (X1 )
    
Xk W
A p (Xk ) LA p (Xk ) SA p (Xk ) HAp (Xk )
    
XK W
A p (XK ) LA p (XK ) SA p (XK ) HAp (XK )

Table 12
Aggregated payoff matrix of choice criteria estimates.

W
D (Xk ) LD (Xk ) SD (Xk ) HD (Xk )

X1 W
D (X1 ) LD (X1 ) SD (X1 ) SD (X1 )
    
Xk W
D (Xk ) LD (Xk ) SD (Xk ) HD (Xk )
    
XK W
D (XK ) LD (XK ) SD (XK ) HD (XK )
max1kK W
D (Xk ) max1kK LD (Xk ) max1kK SD (Xk ) max1kK H
D (Xk )

Table 13
Modied matrix of choice criteria estimates for the rst objective function.

LA p (Xk )

X1 0.67
X2 1.00
X3 0
X4 0.83

Table 14
Modied matrix of choice criteria estimates for the second objective function.

LA p (Xk )

X1 0.25
X2 1.00
X3 0.75
X4 0

The rst stage consists in constructing q payoff matrices (in accordance with the number of objective functions consid-
ered) for all combinations of the given solution alternatives Xk , k = 1, . . . , K and the given representative states of nature
Ys , s = 1, . . . , S. At this point, we do not consider questions of selecting representative states of nature, which are briey
discussed in the next section. To construct payoff matrices it is necessary to resolve S multicriteria problems formalized
52 J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459

Table 15
Aggregated matrix of choice criteria estimates.

LD (Xk )

X1 0.25
X2 1.00
X3 0
X4 0

within the framework of X, M models. By solving them, it is possible to obtain the solution alternatives Xk , k = 1, . . . , K
(with K S). Thereafter, Xk , k = 1, . . . , K are substituted into Fp (X ), p = 1, . . . , q for Ys , s = 1, . . . , S. These substitutions generate
q payoff matrices.
The second stage is related to the analysis of the obtained payoff matrices. The execution of this stage is based on the approach
presented in this paper, namely to consider the choice criteria as objective functions. However, the insucient resolving
capacity of the present stage may lead to non-unique solutions and this circumstance demands to apply the third stage.
The third stage is associated with constructing and analyzing X, R models for the subsequent contraction of decision un-
certainty regions. As it was indicated above, the use of X, R models allows one to take into account indices of quantitative
character as well as qualitative character (based on knowledge, experience, and intuition of involved experts).

The ow chart in Fig. 1 illustrates the general scheme.

4. Application example

The example below details how to perform the rst two stages of the general scheme of multicriteria decision making under
uncertainty.

Fig. 1. Flow chart of the general scheme of multicriteria decision making under information uncertainty.
J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459 53

Consider the following multiobjective problem [33] with interval coecients present in the objective functions:
F1 (x) = [2.70, 3.30]x1 + [11.70, 14.30]x2 + [7.20, 8.80]x3 min; (25)

F2 (x) = [5.40, 6.60]x1 + [3.60, 4.40]x2 + [4.50, 5.50]x3 min (26)


subject to the following constraints:
0 x1 10; (27)

0 x2 12; (28)

0 x3 14; (29)

x1 + x2 + x3 = 30. (30)
The rst stage in the decision making process is associated with building two payoff matrices for all combinations of the
solution alternatives Xk , k = 1, . . . , K and the representative states of nature Ys , s = 1, . . . , S.
We apply LP -sequences to generate the representative states of nature. These sequences have superior characteristics of
uniformity among other uniformly distributed sequences (see [39] for further details and a description of how to construct
LP -sequences).
The results of [39] allow determining points Qs , s = 1, . . . , S with coordinates qst , t = 1, . . . , T in the corresponding unit hy-
percube Q T . Notice that since (25) together with (26) have six coecients, set T = 6. Further, set S = 7. Table 16 presents the
coordinates of Qs , s = 1, . . . , 7, for t = 1, . . . , 6, determined on the basis of [39].
In essence, the selection of representative states of nature is reduced to the selection of points of a uniformly distributed
sequence in Q 6 and their transformation to the hypercube C 6 dened by the lower ct and upper ct bounds of the corresponding
coecients of (25) and (26). Taking this into account, if points Qs , s = 1, . . . , 7 with coordinates qst , t = 1, . . . , 6 form a uniformly
distributed sequence in Q 6 , then points Cs , s = 1, . . . , 7 with the coordinates expressed as
cst = ct + (ct ct )qst , t = 1, . . . , 6 (31)
form a uniformly distributed sequence in C 6
which is included in Table 17.
The coordinates of points in Table 17 serve as a basis for constructing the following seven (in accordance with the number of
the states of nature) multiobjective problems:
F1 (x) = 3.00x1 + 13.00x2 + 8.00x3 min; (32)

F2 (x) = 6.00x1 + 4.00x2 + 5.00x3 min; (33)

F1 (x) = 2.85x1 + 13.65x2 + 7.60x3 min; (34)

Table 16
Points of the -sequences in Q 6 .

s t=1 t=2 t=3 t=4 t=5 t=6

1 0.500 0.500 0.500 0.500 0.500 0.500


2 0.250 0.750 0.250 0.750 0.250 0.750
3 0.750 0.250 0.750 0.250 0.750 0.250
4 0.125 0.625 0.875 0.875 0.625 0.125
5 0.625 0.125 0.375 0.375 0.125 0.625
6 0.375 0.375 0.625 0.125 0.875 0.875
7 0.875 0.875 0.125 0.625 0.375 0.375

Table 17
Representative states of nature.

s t=1 t=2 t=3 t=4 t=5 t=6

1 3.00 13.00 8.00 6.00 4.00 5.00


2 2.85 13.65 7.60 6.30 3.80 5.25
3 3.15 12.35 8.40 5.70 4.20 4.75
4 2.93 12.68 8.20 5.55 4.30 5.38
5 2.78 13.33 8.60 6.45 4.10 4.63
6 3.08 12.03 7.80 5.85 3.70 5.13
7 3.23 13.98 7.40 6.15 3.90 4.88
54 J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459

F2 (x) = 6.30x1 + 3.80x2 + 5.25x3 min; (35)

F1 (x) = 3.15x1 + 12.35x2 + 8.40x3 min; (36)

F2 (x) = 5.70x1 + 4.20x2 + 4.75x3 min; (37)

F1 (x) = 2.93x1 + 12.68x2 + 8.20x3 min; (38)

F2 (x) = 5.55x1 + 4.30x2 + 5.38x3 min; (39)

F1 (x) = 2.78x1 + 13.33x2 + 8.60x3 min; (40)

F2 (x) = 6.45x1 + 4.10x2 + 4.63x3 min; (41)

F1 (x) = 3.08x1 + 12.03x2 + 7.80x3 min; (42)

F2 (x) = 5.85x1 + 3.70x2 + 5.13x3 min; (43)

F1 (x) = 3.23x1 + 13.98x2 + 7.40x3 min; (44)

F2 (x) = 6.15x1 + 3.90x2 + 4.88x3 min (45)


which are subject to the same constraints given in (27)(30).
Using the BellmanZadeh approach to decision making in a fuzzy environment for analyzing X, M models, one can obtain
the following solutions for each problem:

s=1: x01 = 7.00, x02 = 9.00, x03 = 14.00 for (32) and (33);
s=2: x01 = 8.95, x02 = 10.50, x03 = 10.55 for (34) and (35);
s=3: x01 = 7.00, x02 = 9.00, x03 = 14.00 for (36) and (37);
s=4: x01 = 9.95, x02 = 10.50, x03 = 9.55 for (38) and (39);
s=5: x01 = 7.00, x02 = 9.00, x03 = 14.00 for (40) and (41);
s=6: x01 = 9.93, x02 = 11.35, x03 = 8.72 for (42) and (43);
s = 7: x01 = 7.00, x02 = 9.00, x03 = 14.00 for (44) and (45).

Therefore, we can form the following four solution alternatives for the problem (25)(30): X1 = (7.00, 9.00, 14.00), X2 =
(8.95, 10.50, 10.55), X3 = (9.95, 10.50, 9.55), X4 = (9.93, 11.35, 8.72). Substituting them into (32), (34), (36), (38), (40), (42), and
(44), we can construct the payoff matrix for the rst objective function in Table 18. On the other hand, when substituting them
into (33), (35), (37), (39), (41), (44), and (45), we can construct the payoff matrix for the second objective function in Table 20.
The solution of the monocriteria problem (25), subject to (27)(30), based on analyzing the payoff matrix given in Table 18,
generates the corresponding matrix of the choice criteria estimates presented in Table 19.
If one uses the four choice criteria (Wald, Laplace, Savage, and Hurwicz) to analyze the data in Table 19, it follows that all choice
criteria lead to the same solution, i.e., X W = X L = X S = X H = {X3 } (for the choice criterion of Hurwicz, = 0.75 as recommended
in [3]). Thus, the solution alternative may be selected as the solution of the monocriteria problem (25), subject to (27)(30), with
a high degree of condence.
The solution of the monocriteria problem (26), subject to (27)(30), based on analyzing the payoff matrix given in Table 20,
leads to the matrix of choice criteria estimates presented in Table 21.
Performing the same analysis for Table 21, one obtains: X W = {X4 }, X L = {X1 }, X S = {X1 }, and X H = {X4 }. Unlike the previous
case, the solution alternatives X1 and X4 for the second objective function should be considered. Formally, these alternatives
cannot be distinguished on the basis of information given in the payoff matrix of Table 20.

Table 18
Payoff matrix for the rst objective function.

Y1 Y2 Y3 Y4 Y5 Y6 Y7

X1 250.00 249.20 250.80 249.43 259.83 239.03 252.03


X2 243.75 249.01 246.49 245.87 255.58 236.17 253.77
X3 242.75 244.26 241.24 240.60 249.76 231.45 249.60
X4 247.10 249.50 244.70 244.52 253.89 235.14 255.27
J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459 55

Table 19
Matrix of choice criteria estimates for the rst objective function.

F W (Xk ) F L (Xk ) F S (Xk ) F H (Xk )

X1 259.83 250.05 10.07 254.63


X2 255.58 247.81 5.82 250.73
X3 249.76 242.81 0.00 245.18
X4 255.27 247.16 5.24 250.24
249.76 242.81 0.00 245.18
259.83 250.05 10.07 254.63

Table 20
Payoff matrix for the second objective function.

Y1 Y2 Y3 Y4 Y5 Y6 Y7

X1 148.00 151.80 144.20 152.87 146.87 146.07 146.47


X2 148.45 151.67 145.23 151.58 149.62 145.33 147.48
X3 149.45 152.72 146.18 151.75 151.44 146.05 148.75
X4 148.58 151.47 145.69 150.83 150.96 144.82 147.89

Table 21
Matrix of choice criteria estimates for the second objective function.

F W (Xk ) F L (Xk ) F S (Xk ) F H (Xk )

X1 152.87 148.04 2.04 150.70


X2 151.67 148.48 2.75 150.06
X3 152.72 149.48 4.57 151.05
X4 151.47 148.61 4.09 149.81
151.47 148.04 2.04 149.81
152.87 149.48 4.57 151.05

Recall the problem described by (25)(30). Notice that the information in Table 19 permits one to construct the modied
matrix of the choice criteria estimates for the rst objective function given in Table 22. The modied matrix of the choice criteria
for the second objective function, presented in Table 23, was obtained on the basis of information given in Table 21.
The modied matrices of the choice criteria estimates result in the construction of the aggregated payoff matrix presented in
Table 24.
Notice that in this case, the use of the choice criteria of Wald, Laplace, and Hurwicz leads to the same solution alternative:
X W = X L = X H = {X4 }. On the other hand, the criterion of Savage permits one to nd X S = {X2 }. Thus, the solution alternatives
X2 and X4 are the possible results when carrying out the second stage of the general scheme of multicriteria decision making
under information uncertainty. Notice that if the analysis is performed using the generalization of the classic approach proposed
in [15,16], the solution alternatives are X3 and X4 [33]. It is evident that X2 and X4 should be analyzed under the use of < X, R >
models, i.e., the third stage of the general scheme of multicriteria decision making under information uncertainty proposed.

5. Practical applications

The following power engineering problems have been solved with the use of the proposed methodology:

multicriteria power and energy shortage allocation as applied to load management in power systems and subsystems;
multicriteria optimization of network conguration in distribution systems.

Table 22
Modied matrix of choice criteria estimates for the rst objective function.

W
A1 (Xk ) LA1 (Xk ) SA1 (Xk ) HA1 (Xk )

X1 0.00 0.00 0.00 0.00


X2 0.42 0.31 0.42 0.41
X3 1.00 1.00 1.00 1.00
X4 0.45 0.40 0.48 0.46
56 J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459

Table 23
Modied matrix of choice criteria estimates for the second objective function.

W
A2 (Xk ) LA2 (Xk ) SA2 (Xk ) HA2 (Xk )

X1 0.00 1.00 1.00 0.28


X2 0.86 0.69 0.72 0.80
X3 0.11 0.00 0.00 0.00
X4 1.00 0.60 0.19 1.00

Table 24
Aggregated payoff matrix of choice criteria estimates.

W
D (Xk ) LD (Xk ) SD (Xk ) HD (Xk )

X1 0.00 0.00 0.00 0.00


X2 0.42 0.31 0.42 0.41
X3 0.11 0.00 0.00 0.00
X4 0.45 0.40 0.19 0.46

As an example, consider the problem of power and energy shortage allocation. Its solution in [4] is based on multicriteria
allocation of resources or their shortages among consumers (these problems are equivalent from the substantial, mathematical,
and informational points of view). This permits one to consider and minimize diverse implications of power and energy shortage
allocation and to create incentive inuences for consumers [4].
The problem statement supposes the possibility to use diverse types of objective functions dened in a feasible region


n
L= X Rn |0 xi Ai , xi = A (46)
i=1

where X = (x1 , . . . , xn ) is a vector of limitations (for the sake of our considerations) for consumers, Ai is the permissible value of
limitation for the ith consumer, while A is a total value of limitations for all consumers considered in this planning problem.
The general analysis of problems of power and energy shortage allocation, systems of economics management, including
taxation policy, as well as available reported information, has permitted the construction of a general set of goals to solve these
problems in the multicriteria statement. The complete list includes 17 types of goals. Some of them, used in this example, are
given below:

1. Primary limitation of consumers with a lower cost of production or given services on consumed 1 kW h of energy (achieve-
ment of a minimal drop in total production or given services);
2. Primary limitation of consumers with a higher level of the coecient of energy possession of work on consumed 1 kW h
of energy (achievement of maximal reduction in the number of workers, whose productivity and, consequently, wage is
diminished);
3. Primary limitation of consumers with a lower value of the demand coecient (primary limitation of consumers with greater
possibilities of production out the peak time);
4. Primary limitation of consumers with a lower duration of using maximum load in twenty-four hours (primary limitation of
consumers with greater possibilities in transferring maximum load in the daily interval).

The consideration of the indicated goals in one problem with six consumers is associated with minimizing objective func-
tions


6
Fp (X ) = c pi xi , p = 1, 3, 4 (47)
i=1

Table 25
Initial information.

i 1 2 3 4 5 6

c1,i , monetary units/kW h [1.05, 1.95] [2.87, 5.33] [0.98, 1.82] [1.54, 2.86] [0.84, 1.56] [1.49, 2.77]
c2,i [3.78, 7.02] [4.34, 8.06] [4.06, 7.54] [3.71, 6.89] [2.94, 5.46] [3.29, 6.11]
c3,i [0.44, 0.82] [0.23, 0.43] [0.20, 0.36] [0.15, 0.27] [0.18, 0.34] [0.25, 0.47]
c4,i , hours [10.71, 19.89] [12.04, 22.36] [14.77, 27.43] [12.95, 24.05] [12.18, 22.62] [13.72, 25.48]
J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459 57

Table 26
Solution alternatives.

i 1 2 3 4 5 6

X1 8686.60 5999.90 4000.00 50.40 18264.50 12998.60


X2 10021.30 4859.50 4000.00 6974.30 18999.90 5145.00
X3 9440.70 2685.20 3129.40 7000.00 13744.70 14000.00
X4 11685.10 6000.00 4000.00 7000.00 19000.00 2314.90
X5 9876.50 4278.70 4000.00 7000.00 18833.10 6011.70
X6 10603.90 3208.30 4000.00 7000.00 19000.00 6187.80
X7 8000.10 5999.90 4000.00 2707.00 19000.00 10293.00
X8 8543.50 5455.30 2600.00 1053.40 18444.30 13903.50
X9 9162.80 1419.90 3843.70 3166.90 19000.00 13406.70
X10 13031.10 5105.70 4000.00 7000.00 19000.00 1863.20
X11 6782.10 5969.00 4000.00 7000.00 17490.30 8758.60
X12 10475.90 4497.20 4000.00 2679.20 18999.80 9347.90
X13 8159.40 4855.00 2170.80 1818.40 19000.00 13996.40

Table 27
Aggregated matrix of the choice criteria estimates.

i W
D (Xk ) LD (Xk ) SD (Xk ) HD (Xk )

X1 0.00 0.10 0.24 0.06


X2 0.50 0.50 0.06 0.51
X3 0.05 0.00 0.33 0.00
X4 0.22 0.24 0.17 0.24
X5 0.49 0.49 0.03 0.50
X6 0.33 0.40 0.00 0.36
X7 0.11 0.22 0.24 0.14
X8 0.00 0.15 0.16 0.04
X9 0.00 0.00 0.00 0.03
X10 0.00 0.00 0.00 0.00
X11 0.08 0.00 0.00 0.00
X12 0.09 0.05 0.20 0.05
X13 0.07 0.17 0.10 0.00

and maximizing an objective function



6
F2 (X ) = c2i xi (48)
i=1

taking into account (46) with A1 = 16,000 kW, A2 = 6000 kW, A3 = 4000 kW, A4 = 7000 kW, A5 = 19,000, A6 = 7000 kW, and
A = 50,000 kW. In (46)(48), xi , i = 1, . . . , 6 are limitations of power supply for consumers.
The coecients c pi , p = 1, 2, 3, 4, i = 1, . . . , 6 are determined by specic characteristics of the consumers and are given in
Table 25 by the corresponding intervals.
For the problem solution, S = 13 representative states of nature have been elaborated applying the LP -sequences. Table 26
presents the solution alternatives for Ys , s = 1, . . . , 13. Table 27 shows the aggregated payoff matrix of the choice criteria estimates.
The use of the choice criteria of Wald, Laplace, and Hurwicz leads to the same solution: X W = X L = X H = {X2 }. On the other hand,
the criterion of Savage nds X S = {X3 }. Thus, the solution alternatives X2 and X3 (if the analysis is based on the generalization of
the classic approach of [15,16], the solution alternatives are X2 and X5 ) have to be evaluated through the third stage of the general
scheme of multicriteria decision making under uncertainty. However, it is worth noting that applying the rst two stages of the
general scheme, one can rationally reduce the number of solution alternatives from 13 to 2.

6. Conclusions

In this paper, we have presented the reasoning for modifying the generalization, originally proposed in [16], of the classic
approach to deal with information uncertainty in monocriteria decision making for multicriteria problems. This modication
is based on a methodology which permits one to simultaneously consider the choice criteria of the classic approach to taking
into account information uncertainty as objective functions in a multiobjective model framework (X, M models). The main
contribution of this modication is to avoid obtaining contradictory solutions not belonging to the Pareto set (the elaboration of
such solutions violates the basic concept of multicriteria decision making).
A general scheme of multicriteria decision making under information uncertainty has been proposed. This scheme
presents three stages: the rst one consists in building payoff matrices, which involves the solution of multicriteria problems
58 J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459

formalized in the framework of X, M models; the second stage is related to the proposed modication of the generalization
of the classic approach considering the choice criteria as objective functions; and the third stage (which is performed only if
non-unique solutions are obtained in the second stage) is associated with the construction and analysis of X, R models. This
general scheme allows one to use available quantitative information to reduce the decision uncertainty regions. In the case that
non-unique solutions are not obtained, the scheme presumes the application of qualitative information based on knowledge, ex-
perience, and intuition of involved experts. A simple numeric example illustrates, step by step, the rst two stages of the general
scheme. The general scheme has also been applied in a real-world problem of multicriteria power and energy shortage allocation
as applied to load management in power systems and subsystems.
In the present work, the choice criteria of Wald, Laplace, Savage, and Hurwicz have been analyzed and applied. Taking this
into account, one of the future research directions on multicriteria decision making in conditions of uncertainty is associated
with the use of other choice criteria to provide more exibility in selecting solution procedures that support the solution choice
by the DM.

Acknowledgements

This research was supported by the National Council for Scientic and Technological Development of Brazil (CNPq) Grants
305036/2011-4, 307466/2011-6, and 303812/2014-1 and the Energy Company of Minas Gerais (CEMIG) R&D ANEEL Program
Projects D363 and GT480.

References

[1] C.H. Antunes, L.C. Dias, Editorial: managing uncertainty in decision support models, Eur. J. Oper. Res. 181 (2007) 14251426.
[2] R.E. Bellman, L.A. Zadeh, Decision-making in a fuzzy environment, Manage. Sci. 17 (1970) 141164.
[3] L.S. Belyaev, A Practical Approach to Choosing Alternative Solutions to Complex Optimization Problems under Uncertainty, IIASA, Laxenburg, 1977.
[4] R.C. Berredo, P.Ya. Ekel, J.S.C. Martini, R.M. Palhares, R.O. Parreiras, J.G. Pereira Jr., Decision making in fuzzy environment and multicriteria power engineering
problems, Int. J. Electr. Pow. Energy Syst. 33 (2011) 623632.
[5] L. Canha, P. Ekel, J. Queiroz, F. Schuffner Neto, Models and methods of decision making in fuzzy environment and their applications to power engineering
problems, Numer. Linear Algebra Appl. 14 (2007) 369390.
[6] C.A.C. Coelho, Evolutionary multi-objective optimization: Critical review, in: R. Sarker, M. Mohammadian, X. Yao (Eds.), Evolutionary Optimization, Kluwer
Academic Publishers, Boston, 2005, pp. 117146.
[7] W.K. Grassman, Stochastic Systems for Management, North-Holland, Amsterdam, 1981.
[8] I.N. Durbach, T.J. Stewart, Modeling uncertainty in multi-criteria decision analysis, Eur. J. Oper. Res. 223 (2012) 114.
[9] M. Ehrgott, Multicriteria Optimization, Springer-Verlag, Berlin, 2005.
[10] H.A. Eiself, V. Marianov, Multicriteria decision making under uncertainty: a visual approach, Int. Trans. Oper. Res. 21 (2014) 525540.
[11] P.Ya. Ekel, Approach to decision making in fuzzy environment, Comput. Math. Appl. 37 (1999) 5971.
[12] P.Ya. Ekel, Fuzzy sets and models of decision making, Comput. Math. Appl. 44 (2002) 863875.
[13] P.Ya. Ekel, E.A. Galperin, Box-triangular multiobjective linear programs for resource allocation with application to load management and energy market
problems, Math. Comput. Modell. 37 (2003) 117.
[14] P. Ekel, M. Junges, I. Kokshenev, R. Parreiras, Sensitivity and functionally oriented models for power system planning, operation, and control, Int. J. Electr.
Pow. Energy Syst. 45 (2013) 489500.
[15] P. Ekel, I. Kokshenev, R. Palhares, R. Parreiras, F. Schuffner Neto, Multicriteria analysis based on constructing payoff matrices and applying methods of
decision making in fuzzy environment, Optim. Eng. 12 (2011) 529.
[16] P.Ya. Ekel, J.S.C. Martini, R.M. Palhares, Multicriteria analysis in decision making under information uncertainty, Appl. Math. Comput. 200 (2008) 501516.
[17] P.Ya. Ekel, C.A.P.S. Martins, J.G. Pereira Jr., R.M. Palhares, L.N. Canha, Fuzzy set based multiobjective allocation of resources and its applications, Comput.
Math. Appl. 52 (2006) 197210.
[18] P.Ya. Ekel, F.H. Schuffner Neto, Algorithms of discrete optimization and their application to problems with fuzzy coecients, Inf. Sci. 176 (2006) 28462868.
[19] S. French, Uncertainty and imprecision: modelling and analysis, J. Oper. Res. Soc. 7 (1995) 7079.
[20] H. Gaspars-Wieloch, The use of a modication of the Hurwiczs decision rule in multicriteria decision making under complete uncertainty, Bus. Manage.
Educ. 12 (2014) 283302.
[21] E. Herrera-Viedma, F.J. Cabrerizo, J. Kacprzyk, W. Pedrycz, A review of soft consensus models in a fuzzy environment, Inform. Fusion 17 (2014) 413.
[22] E. Herrera-Viedma, F. Herrera, F. Chiclana, A consensus model for multiperson decision making with different preference structures, IEEE Trans. Syst. Man
Cybern. Part A: Syst. Hum. 32 (2002) 394402.
[23] J.L. Hodges Jr., E.L. Lehmann, The use of previous experience in reaching statistical decisions, Ann. Math. Stat. 23 (1952) 396407.
[24] I. Kokshenev, R. Parreiras, P. Ekel, G. Alves, S. Menicucci, A web-based decision support center for electrical energy companies, IEEE Trans. Fuzzy Syst. 23
(2015) 1628.
[25] D. Kuchta, Choice of the best alternative in case of a continuous set of states of nature application of fuzzy numbers, Fuzzy Optim. Decis. Making 6 (2007)
173178.
[26] H.C. Liao, Z.S. Xu, Multi-criteria decision making with intuitionistic fuzzy PROMETHEE, J. Intell. Fuzzy Syst. 27 (2014) 17031717.
[27] H.C. Liao, Z.S. Xu, X.J. Zeng, Distance and similarity measures for hesitant fuzzy linguistic term sets and their application in multi-criteria decision making,
Inf. Sci. 271 (2014) 125142.
[28] R.D. Luce, H. Raiffa, Games and Decisions, John Wiley & Sons, New York, 1957.
[29] X.Y. Ma, Y.Z. Sun, H.L. Fang, Y. Tian, Scenario-based multiobjective decision-making of optimal access point for wind power transmission corridor in the load
centers, IEEE Trans. Sustain. Energy 4 (2013) 229239.
[30] K.M. Miettinen, Nonlinear Multiobjective Optimization, Kluwer Academic Publishers, Boston, 1999.
[31] V. Pareto, Cours dconomie Politique, Lousanne Rouge, Lousanne, 1886.
[32] R.O. Parreiras, P.Ya. Ekel, J.S.C. Martini, R.M. Palhares, A exible consensus scheme for multicriteria group decision making under linguistic assessments, Inf.
Sci. 180 (2010) 10751089.
[33] W. Pedrycz, P. Ekel, R. Parreiras, Fuzzy Multicriteria Decision-Making: Models, Methods, and Applications, John Wiley & Sons, Chichester, 2011.
[34] W. Pedrycz, F. Gomide, An Introduction to Fuzzy Sets: Analysis and Design, MIT Press, Cambridge, 1998.
[35] H. Raiffa, Decision Analysis, Addison-Wesley, Reading, 1968.
[36] S. Rao, Engineering Optimization, John Wiley & Sons, New York, 1996.
J.G. Pereira Jr. et al. / Information Sciences 324 (2015) 4459 59

[37] B. Rottkemper, K. Fischer, Decision making in humanitarian logistics a multi-objective optimization model for relocating relief goods during disaster
recovery operations, in: T. Comes, F. Fiedrich, S. Fortier, J. Geldermann, T.Muller (Eds.), Proceedings of the 10th International ISCRAM Conference, Baden-
Baden, 2013, pp. 647657.
[38] T. Stewart, Dealing with uncertainties in MCDA, in: J. Figueira, S. Greco, M. Ehrgott (Eds.), Multiple Criteria Decision Analysis State of the Art Annotated
Surveys, International Series in Operations Research and Management Science, Springer, New York, 2005, pp. 445470.
[39] M. Sobol, On the systematic search in a hypercube, SIAM J. Numer. Anal. 16 (1979) 790793.
[40] R.I. Trukhaev, Models of Decision Making in Conditions of Uncertainty, Nauka, Moscow, 1981 (in Russian).
[41] M. Wen, K. Iwamura, Fuzzy facility locationallocation problem under the Hurwicz criterion, Eur. J. Oper. Res. 184 (2008) 627635.
[42] Z. Xu, Uncertain Multiple Attribute Decision Making: Methods and Applications, Springer-Verlag, Berlin, 2014.
[43] R.R. Yager, Fuzzy set methods for uncertainty representation in risky nancial decisions, in: Proceedings of the IEEE/IAFE Conference on Computational
Intelligence for Financial Engineering, New York, 1996, pp. 5965.
[44] H.J. Zimmermann, Fuzzy Set Theory and Its Application, Kluwer Academic Publishers, Boston, 1990.
[45] H. Zimmermann, An application-oriented view of modeling uncertainty, Eur. J. Oper. Res. 122 (2000) 190198.

Вам также может понравиться