interactive fuzzy stochastic two-level integer programming through fractile criterion optimization
TRANSCRIPT
ORI GIN AL PA PER
Interactive fuzzy stochastic two-level integerprogramming through fractile criterion optimization
Masatoshi Sakawa • Hideki Katagiri • Takeshi Matsui
Received: 10 September 2009 / Revised: 25 June 2010 / Accepted: 31 July 2010 /
Published online: 14 August 2010
� Springer-Verlag 2010
Abstract In this paper, we focus on stochastic two-level integer programming
problems with cooperative decision makers. Using the fractile criterion optimization
model in chance constrained programming, the formulated stochastic two-level
integer programming problems are transformed into deterministic ones. Taking into
account vagueness of judgments of the decision makers, we present an interactive
fuzzy programming method to derive a satisfactory solution through interactions
with the upper-level decision maker in consideration of the cooperative relation to
the lower-level decision maker. For solving transformed deterministic problems
efficiently, we also introduce genetic algorithms with double strings for nonlinear
integer programming problems. An illustrative numerical example is provided to
demonstrate the feasibility and efficiency of the proposed method.
Keywords Two-level integer programming � Stochastic programming � Interactive
fuzzy programming � Chance constraints � Fractile criterion optimization � Genetic
algorithms � Double strings
1 Introduction
There exist many approaches for two-level programming problems depending on
situations which the decision makers (DMs) are placed in (Sakawa and Nishizaki
2009). Under the assumption that these DMs do not have motivation to cooperate
mutually, the Stackelberg solution (Shimizu et al. 1997) is adopted as a reasonable
solution for the situation. On the other hand, in the case of a project selection
problem in the administrative office of a company and its autonomous divisions, the
M. Sakawa (&) � H. Katagiri � T. Matsui
Faculty of Engineering, Hiroshima University,
1-4-1, Kagamiyama, Higashi-Hiroshima 739-8527, Japan
e-mail: [email protected]
123
Oper Res Int J (2012) 12:209–227
DOI 10.1007/s12351-010-0085-z
situation that these DMs can cooperate with each other seems natural rather than the
noncooperative situation.
Lai (1996) and Shih et al. (1996) proposed solution concepts for two-level linear
programming problems or multi-level ones such that decisions of DMs in all levels
are sequential and all of the DMs essentially cooperate with each other. In their
methods, the DMs identify membership functions of fuzzy goals for their objective
functions. In addition, the DM at the upper level also specifies those of fuzzy goals
for decision variables. The DM at the lower level solves a fuzzy programming
problem with constraints on fuzzy goals of the DM at the upper level. Although
their methods are interesting from the viewpoint of the introduction of fuzzy
concepts for the easy derivation of satisfactory solutions for DMs, there
unfortunately exists a possibility that their methods may lead undesirable final
solution because of the inconsistency between fuzzy goals of objective functions
and those of decision variables. In order to overcome the problem in their methods,
by eliminating the fuzzy goals for the decision variables, Sakawa et al. have
proposed interactive fuzzy programming for two-level or multi-level linear
programming problems to obtain a satisfactory solution for DMs (Sakawa et al.
1998, 2000). The subsequent works on two-level or multi-level programming have
been developing (Lee 2001; Sakawa and Nishizaki 2002a, b, 2009; Sakawa et al.
2001, 2002; Sinha 2003; Pramanik and Roy 2007; Abo-Sinna and Baky 2007;
Roghanian et al. 2007).
In actual decision making situations, however, we must often make a decision on
the basis of vague information or uncertain data. For such decision making
problems involving uncertainty, there exist two typical approaches: probability-
theoretic approach and fuzzy-theoretic one. Stochastic programming, as an
optimization method based on the probability theory, have been developing in
various ways (Stancu-Minasian 1990), including two stage problems considered by
Dantzig (1955) and chance constrained programming proposed by Charnes et al.
(1959). Fuzzy mathematical programming representing the vagueness in decision
making situations by fuzzy concepts have been studied by many researchers
(Rommelfanger 1996; Sakawa 1993). Fuzzy multiobjective linear programming,
first proposed by Zimmermann (1978), have been also developed by numerous
researchers, and an increasing number of successful applications has been appea-
ring (Sakawa et al. 1987; Luhandula 1987; Zimmermann 1987; Slowinski and
Teghem 1990; Lai and Hwang 1992; Sakawa 1993, 2000, 2001; Slowinski 1998).
In particular, after reformulating stochastic multiobjective linear programming
problems using several models for chance constrained programming, Sakawa et al.
(2002, 2003, 2004) presented an interactive fuzzy satisficing method to derive a
satisficing solution for the DM as a generalization of their previous results (Sakawa
and Yano 1985, 1989, 1990; Sakawa et al. 1987; akawa 1993; Sakawa and Kato
2008).
Genetic algorithms (GA) (Holland 1975), initiated by Holland, his colleagues and
his students at the University of Michigan in the 1970s, as stochastic search
techniques based on the mechanism of natural selection and natural genetics, have
received a great deal of attention regarding their potential as optimization
techniques for solving discrete optimization problems or other hard optimization
210 M. Sakawa et al.
123
problems. Although genetic algorithms were not much known at the beginning, after
the publication of Goldberg’s book (Goldberg 1989), genetic algorithms have
recently attracted considerable attention in a number of fields as a methodology for
optimization, adaptation and learning. As we look at recent applications of genetic
algorithms to optimization problems, especially to various kinds of single-objective
discrete optimization problems and/or to other hard optimization problems, we can
see continuing advances (Michalewicz 1992; Gen and Cheng 1996; Back 1996;
Back et al. 1997; Deb 2001; Coello et al. 2002; Eiben and Smith 2003). Focusing on
multiobjective mathematical problems, Sakawa et al., have been advancing genetic
algorithms to derive satisficing solutions to multiobjective problems (Sakawa 2000,
2001).
Under these circumstances, in this paper, we deal with two-level integer
programming problems with random variable coefficients in both objective
functions and constraints. The main contribution of this paper is to provide a novel
decision making methodolgy including a new model, solution concept and solution
algorithm to deal with more realistic problems in the real world, by simultaneously
considering various concepts such as hierarchy structure, fuzziness, randomness,
integer decision variables and interactive fuzzy programming, while most of
previous papers dealt with either of the concepts or a part of them.
Following the concept of the chance constrained programming, the stochastic
two-level linear programming problems are transformed into deterministic nonlin-
ear integer programming ones through the fractile criterion optimization model or
Kataoka’s model (Kataoka 1963). By considering the fuzziness of human
judgments, we present interactive fuzzy programming for deriving a satisfactory
solution for the decision makers. In the proposed interactive method, after
determining the fuzzy goals of the decision makers at both levels, a satisfactory
solution is derived efficiently by updating the satisfactory degree of the decision
maker at the upper level with considerations of overall satisfactory balance among
all the levels. For solving transformed deterministic problems efficiently, we also
propose genetic algorithms with double strings for nonlinear integer programming
problems.
2 Interactive fuzzy stochastic two-level integer programming
Consider two-level integer programming problems with random variable coeffi-
cients formulated as:
minimizefor DM1
z1ðx1; x2;xÞ ¼ c11ðxÞx1 þ c12ðxÞx2
minimizefor DM2
z2ðx1; x2;xÞ ¼ c21ðxÞx1 þ c22ðxÞx2
subject to A1x1 þ A2x2� bðxÞx1j1 2 f0; 1; . . .; m1j1g; j1 ¼ 1; 2; . . .; n1
x2j2 2 f0; 1; . . .; m2j2g; j2 ¼ 1; 2; . . .; n2
9>>>>>=
>>>>>;
ð1Þ
where x1 is an n1 dimensional integer decision variable column vector for the
decision maker at the upper level (DM1), x2 is an n2 dimensional integer decision
Interactive fuzzy stochastic two-level integer programming 211
123
variable column vector for the decision maker at the lower level (DM2), Aj, j = 1, 2
are m 9 nj coefficient matrices, vlji , l = 1, 2, jl = 1, 2,…, nl are positive integer
values, cljðxÞ, l = 1, 2, j = 1, 2 are nj dimensional Gaussian random variable
row vectors with mean vectors �clj and covariance matrices Vlpq, p = 1, 2, q = 1, 2,
and they are independent of each other, and bðxÞ is a random variable vector
whose joint distribution function is F(�). It should be noted here that the two
objective functions z1 and z2 are those of DM1 and DM2, respectively, and
‘‘minimizefor DM1
’’ and ‘‘minimizefor DM2
’’ mean that DM1 and DM2 are minimizers for their
objective functions.
Since (1) contains random variable coefficients, solution methods for ordinary
mathematical programming problems cannot be applied directly. Consequently, we
first deal with the constraints in (1) as chance constraints (Charnes and Cooper
1959) which mean that the constraints need to be satisfied with a certain probability
(satisficing level) and over. Namely, replacing constraints in (1) by chance
constraints with a satisficing level b, the problem can be transformed as:
minimizefor DM1
z1ðx1; x2;xÞ ¼ c11ðxÞx1 þ c12ðxÞx2
minimizefor DM2
z2ðx1; x2;xÞ ¼ c21ðxÞx1 þ c22ðxÞx2
subject to PrfA1x1 þ A2x2� bðxÞg� bx1j1 2 f0; 1; . . .; m1j1g; j1 ¼ 1; 2; . . .; n1
x2j2 2 f0; 1; . . .; m2j2g; j2 ¼ 1; 2; . . .; n2
9>>>>>=
>>>>>;
: ð2Þ
The first constraint in (2) is rewritten as:
PrfA1x1 þ A2x2� bðxÞg� b
, Fðða11x1 þ a12x2Þ; . . .; ðam1x1 þ am2x2ÞÞ� bð3Þ
where F(�) is the joint distribution function for bðxÞ and aij is the i th row vector of
Aj, j = 1, 2.
Therefore, (2) can be rewritten as:
minimizefor DM1
z1ðx1; x2;xÞ ¼ c11ðxÞx1 þ c12ðxÞx2
minimizefor DM2
z2ðx1; x2;xÞ ¼ c21ðxÞx1 þ c22ðxÞx2
subject to Fðða11x1 þ a12x2Þ; . . .; ðam1x1 þ am2x2ÞÞ� bx1j1 2 f0; 1; . . .; m1j1g; j1 ¼ 1; 2; . . .; n1
x2j2 2 f0; 1; . . .; m2j2g; j2 ¼ 1; 2; . . .; n2
9>>>>>=
>>>>>;
: ð4Þ
In the following, for notational convenience, the feasible region of (4) is denoted
by X.
In addition to the chance constraints, it is now appropriate to consider the fractile
criterion optimization model (Kataoka 1963) for the objective functions of (4).
By adopting the model, permissible levels hl, l = 1, 2 such that the probability
which each objective function is better than hl is greater than or equal to some given
threshold hl under the chance constraints are substituted for the original objective
functions zlðx1; x2;xÞ ¼ cl1ðxÞx1 þ cl2ðxÞx2; l = 1, 2 in (4). As a result, the
problem (4) can be transformed as:
212 M. Sakawa et al.
123
minimizefor DM1
h1
minimizefor DM2
h2
subject to Pr c11ðxÞx1 þ c12ðxÞx2� h1½ � � h1
Pr c21ðxÞx1 þ c22ðxÞx2� h2½ � � h2
ðxT1 ; x
T2 Þ
T 2 X
9>>>>>=
>>>>>;
: ð5Þ
In (5), the first two constraints can be converted as:
Pr cl1ðxÞx1 þ cl2ðxÞx2� hlf g� hl
, Prcl1ðxÞx1 þ cl2ðxÞx2 � ð�cl1x1 þ �cl2x2Þ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞVlðxT
1 ; xT2 Þ
Tq � hl � ð�cl1x1 þ �cl2x2Þ
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞVlðxT
1 ; xT2 Þ
Tq
8><
>:
9>=
>;� hl
,Uhl � ð�cl1x1 þ �cl2x2Þffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞVlðxT
1 ; xT2 Þ
Tq
0
B@
1
CA� hl
where Vl is defined as:
Vl ¼Vl11 Vl12
Vl21 Vl22
� �
:
In the above inequality, U(�) is the probability distribution function of the
standard Gaussian distribution. From the monotonicity of the distribution function,
we can define the inverse function U-1(�) of U(�). Then, the above inequalities are
expressed as:
Uhl � ð�cl1x1 þ �cl2x2Þffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞVlðxT
1 ; xT2 Þ
Tq
0
B@
1
CA� hl
, hl � ð�cl1x1 þ �cl2x2Þffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞVlðxT
1 ; xT2 Þ
Tq �U�1ðhlÞ
, hl�ð�cl1x1 þ �cl2x2Þ þ U�1ðhlÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞVlðxT
1 ; xT2 Þ
Tq
:
Letting Kthetal= U-1(hl) and noting that the equality
hl ¼ ð�cl1x1 þ �cl2x2Þ þ Khl
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞVlðxT
1 ; xT2 Þ
Tq
holds at the minimum of hl, problem (5) is equivalent to the following problem:
minimizefor DM1
f1ðx1; x2Þ ¼ ð�c11x1 þ �c12x2Þ þ Kh1
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞV1ðxT
1 ; xT2 Þ
Tq
minimizefor DM2
f2ðx1; x2Þ ¼ ð�c21x1 þ �c22x2Þ þ Kh2
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞV2ðxT
1 ; xT2 Þ
Tq
subject to ðxT1 ; x
T2 Þ
T 2 X
9>>>=
>>>;
: ð6Þ
Interactive fuzzy stochastic two-level integer programming 213
123
In general, it seems natural that the decision makers have fuzzy goals for their
objective functions when they take fuzziness of human judgments into consider-
ation. For each of the objective functions flðx1; x2Þ, l = 1, 2 in (6), assume that the
decision makers have fuzzy goals such as ‘‘flðx1; x2Þ should be substantially less
than or equal to some specific value.’’ Then, (6) can be rewritten as:
maximizeforDM1
l1ðf1ðx1; x2ÞÞmaximize
forDM2l2ðf2ðx1; x2ÞÞ
subject to ðxT1 ; x
T2 Þ
T 2 X
9>=
>;ð7Þ
where ll(�) is a membership function to quantify a fuzzy goal for the lth objective
function in (6) and it is assumed to be monotonically decreasing.
Although the membership function does not always need to be linear, for the sake
of simplicity, we adopt a linear membership function. To be more specific, if the
DM feels that flðx1; x2Þ should be less than or equal to at least fl,0 and
flðx1; x2Þ� fl;1ð\fl;0Þ is satisfactory, the linear membership function llðflðx1; x2ÞÞis defined as:
llðflðx1; x2ÞÞ ¼1; flðx1; x2Þ\fl;1flðx1;x2Þ�fl;0
fl;1�fl;0; fl;1� flðx1; x2Þ� fl;0
0; flðx1; x2Þ[ fl;0
8><
>:ð8Þ
and it is depicted in Fig. 1.
Zimmermann (1978) suggests a method for assessing the parameters of the
membership function. In his method, parameters fl,1, l = 1, 2 are determined as
f1;1 ¼ f1;min ¼ f1ðx11;min; x
12;minÞ ¼ min
ðxT1;xT
2ÞT2X
f1ðx1; x2Þ
f2;1 ¼ f2;min ¼ f2ðx21;min; x
22;minÞ ¼ min
ðxT1;xT
2ÞT2X
f2ðx1; x2Þ
and parameters fl,0, l = 1, 2 are specified as:
f1;0 ¼ f1ðx21;min; x
22;minÞ; f2;0 ¼ f2ðx1
1;min; x12;minÞ
where ðxl1;min; x
l2;minÞ is an optimal solution to the following problem:
Fig. 1 Linear membershipfunction
214 M. Sakawa et al.
123
minimize flðx1; x2Þ ¼ ð�cl1x1 þ �cl2x2Þ þ Khl
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðxT1 ; x
T2 ÞVlðxT
1 ; xT2 Þ
Tq
subject to ðxT1 ; x
T2 Þ
T 2 X
)
: ð9Þ
Then, by setting the parameters as described above, the linear membership
functions (8) is identified.
To derive an overall satisfactory solution to the membership function maximi-
zation problem (7), we first find the maximizing decision of the fuzzy decision
proposed by Bellman and Zadeh (1970). Namely, the following problem is solved
for obtaining a solution which maximizes the smaller degree of satisfaction between
those of the two decision makers:
maximize minfl1ðf1ðx1; x2ÞÞ; l2ðf2ðx1; x2ÞÞgsubject to ðxT
1 ; xT2 Þ
T 2 X
�
: ð10Þ
Solving problem (10), we can obtain a solution which maximizes the smaller
satisfactory degree between those of both decision makers.
If DM1 is satisfied with the optimal solution ðx�1; x�2Þ to (10), it follows that the
optimal solution ðx�1; x�2Þ becomes a satisfactory solution; however, DM1 is not
always satisfied with the solution ðx�1; x�2Þ. It is quite natural to assume that DM1
specifies the minimal satisfactory level d̂ 2 ð0; 1Þ for the membership function
l1ðf1ðx1; x2ÞÞ subjectively.
Consequently, if DM1 is not satisfied with the solution ðx�1; x�2Þ to problem (10),
the following problem is formulated:
maximize l2ðf2ðx1; x2ÞÞsubject to l1ðf1ðx1; x2ÞÞ� d̂
ðxT1 ; x
T2 Þ
T 2 X
9=
;ð11Þ
where DM2’s membership function l2ðf2ðx1; x2ÞÞ is maximized under the condition
that DM1’s membership function l1ðf1ðx1; x2ÞÞ is larger than or equal to the
minimal satisfactory level d̂ specified by DM1.
If there exists an optimal solution ðx�1; x�2Þ to problem (11), it follows that DM1
obtains a satisfactory solution having a satisfactory degree larger than or equal to
the minimal satisfactory level specified by DM1’s self. However, the larger the
minimal satisfactory level d̂ is assessed, the smaller the DM2’s satisfactory degree
becomes when the membership functions of DM1 and DM2 conflict with each
other. Consequently, a relative difference between the satisfactory degrees of DM1
and DM2 becomes larger, and it follows that the overall satisfactory balance
between both decision makers is not appropriate.
In order to take account of the overall satisfactory balance between both decision
makers, DM1 needs to compromise with DM2 on DM1’s own minimal satisfactory
level. To do so, the following ratio of the satisfactory degree of DM2 to that of DM1
is helpful:
D ¼ l2ðf2ðx1; x2ÞÞl1ðf1ðx1; x2ÞÞ
which is originally introduced by Lai (1996).
Interactive fuzzy stochastic two-level integer programming 215
123
DM1 is guaranteed to have a satisfactory degree larger than or equal to the
minimal satisfactory level for the fuzzy goal because the corresponding constraint is
involved in problem (11). To take into account the overall satisfactory balance
between both decision makers, DM1 specifies the lower bound Dmin and the upper
bound Dmax of the ratio D, and D is evaluated by verifying whether or not it is in the
interval [Dmin, Dmax]. The condition that the overall satisfactory balance is
appropriate is represented by
D 2 ½Dmin;Dmax�:At the iteration k, let ðxk
1; xk2Þ, fk
l ¼ flðxk1; x
k2Þ, llðfk
l Þ and Dk ¼ l2ðfk2Þ=l1ðfk
1Þdenote the current solution, DM1’s objective function value, DM1’s satisfactory
degree and the ratio of satisfactory degrees of the two decision makers, respectively.
The interactive process terminates if the following two conditions are satisfied and
DM1 concludes the solution as an overall satisfactory solution.
2.1 Termination conditions of the interactive process
Condition 1 DM1’s satisfactory degree is larger than or equal to the minimal
satisfactory level d̂ specified by DM1’s self, i.e., l1ðfk1Þ� d̂.
Condition 2 The ratio Dk of satisfactory degrees lies in the closed interval
between the lower and the upper bounds specified by DM1, i.e., Dk 2 ½Dmin;Dmax�.Condition 1 ensures the minimal satisfaction to DM1 in the sense of the
attainment of the fuzzy goal, and condition 2 is provided in order to keep overall
satisfactory balance between both decision makers. If these two conditions are not
satisfied simultaneously, DM1 needs to update the minimal satisfactory level d̂.
The updating procedures are summarized as follows.
2.2 Procedure for updating the minimal satisfactory level d̂
Case 1 If condition 1 is not satisfied, then DM1 decreases the minimal
satisfactory level d̂.
Case 2 If the ratio Dk exceeds its upper bound, then DM1 increases the minimal
satisfactory level d̂. Conversely, if the ratio Dk is below its lower bound, then
DM1 decreases the minimal satisfactory level d̂.
Case 3 Although conditions 1 and 2 are satisfied, if DM1 is not satisfied with the
obtained solution and judges that it is desirable to increase the satisfactory degree
of DM1 at the expense of the satisfactory degree of DM2, then DM1 increases the
minimal satisfactory level d̂. Conversely, if DM1 judges that it is desirable to
increase the satisfactory degree of DM2 at the expense of the satisfactory degree
of DM1, then DM1 decreases the minimal satisfactory level d̂.
In particular, if condition 1 is not satisfied, there does not exist any feasible solution
for problem (11), and therefore DM1 has to moderate the minimal satisfactory level.
Now we are ready to propose interactive fuzzy programming for deriving a
satisfactory solution by updating the satisfactory degree of the decision maker at the
upper level with considerations of overall satisfactory balance among all the levels.
216 M. Sakawa et al.
123
2.3 Computational procedure of interactive fuzzy programming
Step 1: Ask the decision maker at the upper level, DM1, to subjectively determine
a satisficing level b [ (0, 1) for constraints in (2) and probability levels hl, l = 1, 2.
Go to Step 2.
Step 2: Solve (9) and calculate fl,min via optimal solutions ðxl1;min, xl
2;minÞ, l = 1, 2
to (9). Then, identify the linear membership function llðflðx1; x2ÞÞ of the fuzzy
goal for the corresponding objective function using the Zimmermann method
(Zimmermann 1978). Go to Step 3.
Step 3: Set k := 1. Solve the maximin problem (10) for obtaining a solution which
maximizes the smaller degree of satisfaction between those of the two decision
makers and calculate fkl ¼ flðxk
1; xk2Þ, llðfk
l Þ, l = 1, 2 and Dk ¼ l2ðfk2Þ=l1ðfk
1Þ via
the optimal solution ðxk1; x
k2Þ to (10). If DM1 is satisfied with the optimal solution
to (10), the optimal solution becomes a satisfactory solution and this interaction
procedure is terminated. Otherwise, ask DM1 to subjectively set the minimal
satisfactory level d̂ 2 ð0; 1Þ for the membership function l1ðf1ðx1; x2ÞÞ. Further-
more, ask DM1 to set the upper bound Dmax and the lower bound Dmin for D. Go
to Step 4.
Step 4: Set k := k ? 1. Solve problem (11) for finding a solution to maximize
DM2’s membership function l2ðf2ðx1; x2ÞÞ under the condition that DM1’s
membership function l1ðf1ðx1; x2ÞÞ is larger than or equal to the minimal
satisfactory level d̂ and calculate fkl ¼ flðxk
1; xk2Þ, llðfk
l Þ, l = 1, 2 and Dk ¼l2ðfk
2Þ=l1ðfk1Þ via the optimal solution ðxk
1; xk2Þ to (11). Go to Step 5.
Step 5: If the current solution ðxk1; x
k2Þ satisfies the termination conditions and
DM1 accepts it, then the procedure stops and the current solution is determined to
be a satisfactory solution. Otherwise, ask DM1 to update the minimal satisfactory
level d̂, and return to Step 4.
In the proposed interactive fuzzy programming method, it is required to solve the
nonlinear integer programming problems (9), (10) and (11), which is apparently
difficult to solve compared to linear integer programming problems and 0–1
nonlinear programming problems. In order to solve such difficult problems, we
propose genetic algorithms designed for nonlinear integer programming problems,
called Genetic Algorithms with Double Strings for Nonlinear Integer Programming
(GADSNIP). We do not employ the state-of-the-art multiobjective evolutionary
algorithms such as NSGA-II and SPEA2 because they are mainly weighted toward
finding the Pareto optimal solution set so as to cover the whole set, whereas the
goal of this paper is to derive a compromise or satisficing solution for the decision
maker.
As one of the most promising methods for solving combinatorial optimization
problems, tabu search have been developed. From the point of view of our
experience, however, in most situations, the Genetic Algorithm with Double String
(GADS) is better than Tabu Search (TS) methods in the case of nonlinear integer
programming problems, whereas TS methods are probably superior to GA-based
methods in the case of 0–1 decision variables. The reason is related to the two
Interactive fuzzy stochastic two-level integer programming 217
123
computational aspects in metaheuristics; one is the diversity of the solutions
generated from these methods, and the other is the computational cost for finding
the promising direction to obtain better solutions.
As for the first issue, it should be noted here that in general, the diversity of the
solutions explored in integer programming problems should be greater than that in
0–1 programming problems since the feasible region of a 0–1 programming
problem is definitely smaller than that of an integer programming problem if
the objective functions and constraints in both the problems are the same except for
the condition for the taken values of decision variables. Since the diversity of the
solutions generated by GA-based methods is generally greater than that by TS-based
methods, if the range of the integer values taken by decision variables is wider, the
performance of GA-based methods has a tendency to be better than that of TS-based
methods.
Moreover, when using TS-based methods for solving nonlinear problems, the
computational cost for finding the best searching direction clearly becomes far more
expensive compared to the linear cases because the differences of the objective
function values between the adjacent solutions (or moves) are very easily calculated
in the linear cases. Therefore, as the nonlinearity of the problems to be solved
becomes stronger, the degree of superiority of TS-based methods over GA-based
methods decreases.
Considering that the goal of this paper is to solve the problems where the
objective functions are nonlinear and the decision variables are integer, we employ a
GA-based method to derive a satisficing solution for the decision maker.
3 Genetic algorithm with double strings for nonlinear integer programming
As discussed above, in this section, for solving the transformed deterministic
nonlinear integer programming problems efficiently, we propose Genetic Algo-
rithms with Double Strings for Nonlinear Integer Programming (GADSNIP). As an
efficient approximate solution method, GADSNIP are designed for nonlinear integer
programming problems formulated as:
minimize f ðxÞsubject to giðxÞ� 0; i ¼ 1; 2; . . .;m
xj 2 f0; 1; . . .; mjg; j ¼ 1; 2; . . .; n
9=
;ð12Þ
where x is an n dimensional integer decision variable column vector. Furthermore,
f(�) and gi(�), i = 1, 2,…, m may be nonlinear. If mj = 1 for all j [ {1, 2,…, n},
(12) is called a nonlinear 0–1 programming problem.
Quite similar to GADS (Sakawa 2001), an individual is represented by a double
string shown in Table 1. In Table 1, for a certain j, s(j) [ {1, 2,…, n} represents an
Table 1 Double strings(1) s(2) … s(n)
ys(1) ys(2) … ys(n)
218 M. Sakawa et al.
123
index of a decision variable xs(j) in the solution space, while ys(j), j = 1, 2,…, n does
the integer value among {0, 1,…, mj} of the s(j) th decision variable xs(j).
In GADSNIP, as with GADS (Sakawa 2001), a feasible solution used in
decoding, called a reference solution, must be found before the execution of the
genetic algorithm. One possible way to obtain a feasible solution to (12), is to
maximize the exponential function for the violation of constraints defined by:
GðxÞ ¼ exp �hXm
i¼1
MgiðxÞ
�gi
� �" #
ð13Þ
where
MðnÞ ¼ n; n� 0
0; n\0
�
;
�gi’s are parameters for normalization and h is a positive constant. Namely, for
obtaining a feasible solution, solve an unconstrained maximization problem
maximize GðxÞsubject to xj 2 f0; 1; . . .; mjg; j ¼ 1; 2; . . .; n
�
ð14Þ
through GADSNIP (without using the decoding algorithm) by regarding GðxÞ as the
fitness function of an individual S. A solution x0 such as Gðx0Þ ¼ 1 to this problem
is regarded as a reference solution x� :¼ x0.
In the following, we construct the decoding algorithm for GADSNIP using a
reference solution x�, where N is the number of individuals.
3.1 Decoding algorithm using a reference solution
Step 1: Let j := 1, x :¼ 0, l := 0. Go to Step 2.
Step 2: Let xs(j) := ys(j). Go to Step 3.
Step 3: If giðxÞ� 0, i = 1, 2,…, m, let l := j, j := j ? 1 and go to Step 4.
Otherwise, let j := j ? 1 and go to Step 5.
Step 4: If j B n, go to step 2. Otherwise, go to Step 5.
Step 5: If l [ 0, go to Step 6. Otherwise, go to Step 7.
Step 6: For j such that 1 B j B l, let xs(j) := ys(j). Then, for j such that
(l ? 1) B j B n, let xs(j) := 0. Since the resulting x is a feasible solution, quit the
process.
Step 7: Let j := 1, x :¼ x�. Go to Step 8.
Step 8: Let xs(j) := ys(j). If ysðjÞ ¼ x�sðjÞ, let j := j ? 1 and go to Step 10. Otherwise,
if ysðjÞ 6¼ x�sðjÞ, go to Step 9.
Step 9: If giðxÞ� 0, i = 1, 2,…, m, let j := j ? 1 and go to Step 10. Otherwise,
let xsðjÞ :¼ x�sðjÞ, j := j ? 1 and go to Step 10.
Step 10: If j B n, return to Step 8. Otherwise, since the resulting x is a feasible
solution, quit the process.
It is significant to realize that the diversity of solutions x greatly depends on the
reference solution used in the above decoding algorithm. To overcome such
Interactive fuzzy stochastic two-level integer programming 219
123
situations, we propose the following reference solution updating procedure such
that the current reference solution is updating by another feasible solution if the
diversity of solutions seems to be lost. To do so, for every generation, check
the dependence on the reference solution through the calculation of the mean of
the Hamming distance between all solutions decoded from individuals and the
reference solution are checked. If the dependence on the reference solution is
strong, the reference solution is replaced by the solution corresponding to an
individual having maximum Hamming distance. Let N, x�, g(\ 1.0) and xr
respectively denote the number of individuals, the reference solution, a parameter
for reference solution updating and a feasible solution decoded by the r th
individual, then the reference solution updating procedure can be described as
follows.
3.2 Reference solution updating procedure
Step 1: Set r := 1, rmax := 1, dmax := 0 and dsum := 0. Go to Step 2.
Step 2: Calculate dr ¼Pn
j¼1 jxrj � x�j j and let dsum :¼ dsum þ dr. If dr [ dmax and
f ðxrÞ\f ðx�Þ, let dmax := dr, rmax := r and r := r ? 1, and go to Step 3.
Otherwise, let r := r ? 1 and go to Step 3.
Step 3: If r [ n, go to Step 4. Otherwise, return to Step 2.
Step 4: If dsum=ðN �Pn
j¼1 mjÞ\g, then update the reference solution as x� :¼ xrmax ,
and stop. Otherwise, stop without updating the reference solution.
It should be observed here that when the constraints of the problem are strict, there
exist a possibility that all of the individuals are decoded in the neighborhood of the
reference solution. To avoid such a possibility, in addition to the reference solution
updating procedure, after every P generations, the reference solution is replaced by
another feasible solution.
In the proposed method, we determine values of individuals of the initial
population by using the property that the optimal solution to an integer
programming problem is close to that to the corresponding continuous relaxation
problem. To be more specific, when we determine initial values of the lower string
of each individual, ys(j), j = 1, 2,…, n, we use Gaussian random numbers with mean
x̂j and variance r2. Here x̂j is the j th element of the optimal solution to the following
continuous relaxation problem of (12).
minimize f ðxÞsubject to giðxÞ� 0; i ¼ 1; 2; . . .;m
0� xj� mj; j ¼ 1; 2; . . .; n
9=
;ð15Þ
If (15) is a convex programming problem, we can obtain an optimal solution by an
existing convex programming technique. Otherwise, we find an approximate optimal
solution by using some nonconvex programming technique, e.g., GENOCOP V
(Kozieł and Michalewicz 1999), a genetic algorithm for general nonlinear program-
ming problems.
220 M. Sakawa et al.
123
3.3 Generation of initial population by using continuous relaxation
Step 1: Let r := 1. Go to Step 2.
Step 2: Generate a uniform random number rand() [ [0,1). If it is less than or
equal to the prefixed parameter R, go to Step 3. Otherwise, go to Step 7.
Step 3: Let j := 1. Go to Step 4.
Step 4: Select a number from among {1, 2,…, n} randomly, and let it s(j). Note
that s(j) must not be equal to s(j0), j0 = 1, 2,…, j - 1. Go to Step 5.
Step 5: For the current s(j), using the Gaussian integer random number with mean
x̂sðjÞ and variance r2, determine the value of ys(j) as ysðjÞ :¼ Gaussðx̂sðjÞ; r2Þ. Let
j := j ? 1 and go to Step 6.
Step 6: If j [ n, let r := r ? 1 and go to Step 11. Otherwise, return to Step 4.
Step 7: Let j := 1. Go to Step 8.
Step 8: Select a number from among {1, 2,…, n} randomly, and let it s(j). Note
that s(j) must not be equal to s(j0), j0 = 1, 2,…, j - 1. Go to Step 9.
Step 9: For the current s(j), determine the value of ys(j) by using the uniform
integer random number on the set {0, 1,…, ms(j)}, and let j := j ? 1. Go to
Step 10.
Step 10: If j [ n, let r := r ? 1 and go to Step 11. Otherwise, return to Step 8.
Step 11: If r [ N, quit this process. Otherwise, return to Step 2.
As a reproduction operator, elitist expected value selection, which is the
combination of expected value selection and elitist preserving selection, is adopted.
In Sakawa et al. 2005, elitist expected value selection is defined as a combination of
elitism and expected value selection as mentioned below.
Elitism: If the fitness of the best individual Sbest (elite) in all past populations is
less than that of the best individual S� in the current population, preserve it as the
elite, i.e., Sbest :¼ S�. Otherwise, incorporate Sbest into the current population.
Expected value selection: For a population consisting of N individuals, the expected
value of the number of each individual Sr in the next population
Nr ¼f ðSrÞ
PNr¼1 f ðSrÞ
� N
is calculated. Then, the integral part of Nr means the deterministic number of Sr
preserved in the next population. While, the decimal part of Nr is regarded as the
probability that one Sr can survive, i.e., N �PN
r¼1 Nr individuals are determined on
the basis of this probability.
If either the single-point crossover or the multi-point crossover is directly applied
to individuals of double string type, the k th element of an offspring may take the
same number that the k0 th element takes. Similar violation occurs in solving
traveling salesman problems or scheduling problems through genetic algorithms as
well. In order to avoid this violation, a crossover method called partially matched
crossover (PMX) was proposed (Goldberg and Lingle 1985) PMX suitable for
double strings can be constructed as follows.
Interactive fuzzy stochastic two-level integer programming 221
123
3.4 PMX for double string
Step 1: Select two individuals X, Y from the population as parent individuals and
prepare copies X0 and Y0 of X and Y, respectively. Go to Step 2.
Step 2: Choose two crossover points at random on these strings, say, h and
k(h \ k). Go to Step 3.
Step 3:
(a) Set j = h.
(b) Find j0 such that sX’(j0) = sY(j).
Then, interchange ðsX0 ðjÞ; ysX0 ðjÞÞT
with ðsX0 ðj0Þ; ysX0 ðj0ÞÞT
and
set j = j ? 1.
(c) If j [ k, stop. Otherwise, go to (b).
Go to Step 4.
Step 4: Replace the part from h to k of X0 with that of Y and let X0 be the offspring
of X. Go to step 5.
Step 5:
(a) Set j = h.
(b) Find j0 such that sY’(j0) = sX(j).
Then, interchange ðsY 0 ðjÞ; ysY 0 ðjÞÞT
with ðsY 0 ðj0Þ; ysY 0 ðj0ÞÞT
and
set j = j ? 1.
(c) If j [ k, stop. Otherwise, return to (b). Go to Step 6.
Step 6: Replace the part from h to k of Y0 with that of X and let Y0 be the offspring
of Y.
It is considered that mutation plays the role of local random search in genetic
algorithms. In this paper, two mutation operators (bit-reverse type and inversion) are
used. A direct extension of mutation for 0–1 programming problems is to change the
value of ys(j) at random in [0,ms(j)] uniformly, when mutation occurs at ys(j). The
mutation operator is further refined by using the information about the solution of
the continuous relaxation problem x̂. To be more explicit, the following algorithm is
carried out.
3.5 Mutation of bit reverse type for double strings
Step 1: Set r := 1.
Step 2: Set j := 1.
Step 3: Generate a uniform random number rand() [ [0,1). If it is less than or
equal to the given mutation rate pm, then go to step 4. Otherwise, go to step 7.
Step 4: Generate a uniform random number rand() [ [0,1) again. If it is less than
or equal to the prefixed parameter R, go to step 5. Otherwise, go to step 6.
Step 5: Using the Gaussian integer random number with mean x̂sðjÞ and variance
s2, determine the value of ys(j) as ysðjÞ :¼ Gaussðx̂sðjÞ; s2Þ and go to step 7.
222 M. Sakawa et al.
123
Step 6: Determine the value of ys(j) by using the uniform integer random number
on the set {0, 1,…, ms(j)} and go to step 7.
Step 7: If j \ n, set j := j ? 1 and return to step 3 Otherwise, go to step 8.
Step 8: If r \ N, set r := r ? 1 and return to step 2. Otherwise, stop.
3.6 Inversion operator
Step 1: After determining the two inversion points h and k (h \ k), pick out a part
of the upper row of a double string from h to k.
Step 2: Arrange the substring in the reverse order.
Step 3: Put the arranged substring back in the double string.
From the results discussed thus far, it is now appropriate to present the following
computational procedures of GADSNIP.
3.7 Computational procedures of GADSNIP
Step 1: Determine values of the parameters used in the genetic algorithm: the
population size N, the generation gap G, the crossover rate pc, the mutation rate pm,
the inversion rate pi, the minimal search generation number Imin, the maximal
search generation number Imax([ Imin), the scaling constant cmult, the convergence
criterion e, the degree of use of information about solutions to linear programming
relaxation problems R, the parameter for reference solution updating g and the
penalty constant h. Set the generation counter t at 0. Go to Step 2.
Step 2: Generate the initial population consisting of N individuals based on the
information of the optimal solution to the continuous relaxation problem. Go to
Step 3.
Step 3: Decode each individual in the current population and calculate its fitness
based on the corresponding solution. Go to Step 4.
Step 4: If the termination condition is fulfilled, stop. Otherwise, let t := t ? 1 and
go to Step 5.
Step 5: Apply reproduction operator using elitist expected value selection after
linear scaling. Go to Step 6.
Step 6: Apply crossover operator, called PMX (Partially Matched Crossover) for
double string. Go to Step 7.
Step 7: Apply mutation based on the information of a solution to the continuous
relaxation problem. Go to Step 8.
Step 8: Apply inversion operator. return to Step 3.
4 Numerical example
To demonstrate the feasibility and efficiency of the proposed method, consider the
following two-level integer programming problem involving random variable
coefficients.
Interactive fuzzy stochastic two-level integer programming 223
123
minimizefor DM1
z1ðx1; x2;xÞ ¼ c11ðxÞx1 þ c12ðxÞx2
minimizefor DM2
z2ðx1; x2;xÞ ¼ c21ðxÞx1 þ c22ðxÞx2
subject to a11x1 þ a12x2� b1ðxÞa21x1 þ a22x2� b2ðxÞ
..
.
a101x1 þ a102x2� b10ðxÞx1j1 2 f0; 1; . . .; 30g; j1 ¼ 1; . . .; 15
x2j2 2 f0; 1; . . .; 30g; j2 ¼ 1; . . .; 15
9>>>>>>>>>>>>=
>>>>>>>>>>>>;
ð16Þ
where x1 ¼ ðx11; . . .; x115ÞT , x2 ¼ ðx21; . . .; x215ÞT , bðxÞ is also a Gaussian random
vector whose mean is given as:
�b ¼ ð4776; 2790; 5082;�1116; 4944; 2160; 3390; 1608; 3408; 3540Þ
and clðxÞ ¼ ðcl1ðxÞ; cl2ðxÞÞ, l = 1, 2 are Gaussian random vectors whose mean are
given as Table 2. Upper bounds of all decision variables are equal to 30.
Parameters of GADSNIP are set as: population size N = 100, generation gap
G = 0.9, crossover rate pc = 0.9, mutation rate pm = 0.05, inversion rate
pi = 0.05, minimal search generation number Imin = 500, maximal search gener-
ation number Imax = 1000, scaling constant cmult = 1.6, convergence criterion
e = 0.01, degree of use of information about solutions to linear programming
relaxation problems R = 0.9, parameter for reference solution updating g = 0.2,
penalty constant h = 5.
Suppose that DM1 specifies a satisficing level b and probability levels hl, l = 1, 2
as b = 0.70, h1 = 0.75, h2 = 0.70. Using GADSNIP minimal values fl,min of
objective functions flðx1; x2Þ under the chance constraints corresponding to the
satisficing level b are calculated as f1;min ¼ f1ðx11;min; x
12;minÞ ¼ �1080:02,
f2;min ¼ f2ðx21;min; x
22;minÞ ¼ �508:68.
Suppose that the DMs employ the linear membership function (8) whose
parameters are determined by the Zimmermann method (Zimmermann 1978). Then,
parameter values fl,1 and fl,0, l = 1, 2 characterizing membership functions ll(�) are
becomes:
f1;1 ¼ �1080:02; f1;0 ¼ 628:54; f2;1 ¼ �508:68; f2;0 ¼ 1251:00:
The maximin problem is solved by GADSNIP and the obtained result is shown at
the column labeled ‘‘1st’’ in Table 3. Since DM1 is not satisfied with the current
solution, DM1 sets the minimal satisfactory level d̂ 2 ð0; 1Þ for l1ðf1ðx1; x2ÞÞ to
0.70 so that l1ðf1ðx1; x2ÞÞ will be improved from its current value. Furthermore, the
Table 2 Means of coefficients of objective functions
�c11 9 -1 10 -6 4 -6 3 -9 10 -5 7 -4 8 -9 3
�c12 -9 5 -8 2 -10 3 -2 4 -3 5 -2 9 -9 6 -6
�c21 10 2 -5 7 2 -3 -2 7 7 9 8 -5 6 7 9
�c22 4 2 -6 6 3 4 -2 9 -3 -10 2 8 -7 8 -9
224 M. Sakawa et al.
123
upper bound and the lower bound of the ratio of satisfactory degrees D are set as
Dmax = 0.80 and Dmin = 0.70.
Problem (11) for d̂ ¼ 0:70 is solved using GADSNIP, and the obtained result of
the second iteration is shown at the column labeled ‘‘2nd’’ in Table 3.
Since the ratio of satisfactory degrees D2 exceeds Dmax = 0.80, the second
condition of termination of the interactive process is not fulfilled. Suppose that DM1
feels that l1 should be considerably better than l2, and DM1 updates the minimal
satisfactory level d̂ from 0.70 to 0.80 in order to improve l1. Consequently, problem
(11) for d̂ ¼ 0:80 is solved by using GADSNIP, and the obtained result is shown at
the column labeled ‘‘3rd’’ in Table 3.
Since the ratio of satisfactory degrees D3 is less than Dmin = 0.70, the second
condition of termination of the interactive process is not fulfilled. Hence, DM1
updates the minimal satisfactory level d̂ from 0.80 to 0.75 for improving l2 at the
sacrifice of l1. As a result, problem (11) for d̂ ¼ 0:75 is solved by GADSNIP, and
the obtained result is shown at the column labeled ‘‘4th’’ in Table 3.
In this example, at the fourth iteration, since the current solution satisfies all
termination conditions of the interactive process and DM1 is satisfied with the
current solution, the satisfactory solution is obtained.
5 Conclusions
In this paper, we focused on two-level integer programming problems involving
random variable coefficients. Using the fractile criterion optimization model in
chance constrained programming, the formulated stochastic two-level integer
programming problems were transformed into deterministic ones. Taking into
account vagueness of judgments of the decision makers, and the assumption that
the upper-level decision maker and the lower-level one in the problem are
mutually cooperative, interactive fuzzy programming has been presented. In the
proposed interactive method, after determining the fuzzy goals of the decision
makers at both levels, a satisfactory solution can be derived efficiently by
updating the satisfactory degree of the decision maker at the upper level with
considerations of overall satisfactory balance among all the levels. Genetic
algorithms designed for nonlinear integer programming problems, called Genetic
Table 3 Interaction process
Interaction 1st 2nd 3rd 4th
d̂ – 0.70 0.80 0.75
f1ðxk1; x
k2Þ -499.65 -571.51 -750.63 -653.35
f2ðxk1; x
k2Þ 85.22 149.97 318.24 255.69
l1ðf1ðxk1; x
k2ÞÞ 0.660313 0.702376 0.807210 0.750273
l2ðf2ðxk1; x
k2ÞÞ 0.662494 0.625696 0.530071 0.565616
Dk 1.003306 0.890828 0.656671 0.753880
Interactive fuzzy stochastic two-level integer programming 225
123
Algorithm with Double Strings for Nonlinear Integer Programming (GADSNIP),
were also introduced for solving transformed deterministic ones efficiently.
An illustrative numerical example was provided to demonstrate the feasibility and
efficiency of the proposed method. Extensions to two-level integer programming
problems involving fuzzy random variable coefficients will be required in the near
future.
References
Abo-Sinna MA, Baky IA (2007) Interactive balance space approach for solving multi-level multi-
objective programming problems. Inf Sci 177:3397–3410
Back T (1996) Evolutionary algorithms in theory and practice. Oxford University Press, New York
Back T, Fogel DB, Michalewicz Z (1997) Handbook of evolutionary computation. Institute of Physics
Publishers, Bristol
Bellman RE, Zadeh LA (1970) Decision making in a fuzzy environment. Manag Sci 17:141–164
Charnes A, Cooper WW (1959) Chance constrained programming. Manag Sci 6:73–79
Coello CAC, Van Veldhuizen DA, Lamont GB (2002) Evolutionary algorithms for solving multi-
objective problems. Kluwer, New York
Dantzig GB (1955) Linear programming under uncertainty. Manag Sci 1:197–206
Deb K (2001) Multi-objective optimization using evolutionary algorithms. Wiley, Chichester
Eiben AE, Smith JE (2003) Introduction to evolutionary computing. Springer, Berlin
Gen M, Cheng R (1996) Genetic algorithms and engineering design. Wiley, New York
Goldberg DE, Lingle R (1985) Alleles, loci, and the traveling salesman problem. In: Proceedings of 1st
international conference on genetic algorithms and their applications, pp 154–159
Goldberg DE (1989) Genetic algorithms in search, optimization, and machine learning. Addison Wesley,
Massachusetts
Holland JH (1975) Adaptation in natural and artificial systems. University of Michigan Press, Ann Arbor
Kataoka S (1963) A stochastic programming model. Econometrica 31:181–196
Kozieł S, Michalewicz Z (1999) Evolutionary algorithms, homomorphous mapping, and constrained
parameter optimization. Evol Comput 7:19–44
Lai YJ (1996) Hierarchical optimization: a satisfactory solution. Fuzzy Sets Syst 77:321–335
Lai YJ, Hwang CL (1992) Fuzzy mathematical programming. Springer, Berlin
Lee ES (2001) Fuzzy multiple level programming. Appl Math Comput 120:79–90
Luhandjula MK (1987) Multiple objective programming problems with possibilistic coefficients. Fuzzy
Sets Syst 21:135–145
Michalewicz Z (1992) Genetic algorithms ? data structures = evolution programs. Springer, Berlin
(1992). Second extended edition (1994). Third revised and extended edition (1996)
Pramanik S, Roy TK (2007) Fuzzy goal programming approach to multilevel programming problems. Eur
J Oper Res 176:1151–1166
Roghanian E, Sadjadi SJ, Aryanezhad MB (2007) A probabilistic bi-level linear multi-objective
programming problem to supply chain planning. Appl Math Comput 188:786–800
Rommelfanger H (1996) Fuzzy linear programming and applications. Eur J Oper Res 92:512–527
Sakawa M (1993) Fuzzy sets and interactive multiobjective optimization. Plenum Press, New York
Sakawa M (2000) Large scale interactive fuzzy multiobjective programming. Physica-Verlag, Heidelberg
Sakawa M (2001) Genetic algorithms and fuzzy multiobjective optimization. Kluwer, Boston
Sakawa M, Kato K (2008) Interactive fuzzy multi-objective stochastic linear programming. In: Kahraman
C (ed) Fuzzy multi-criteria decision making—theory and applications with recent developments.
Springer, New York, pp 375–408
Sakawa M, Nishizaki I (2002a) Interactive fuzzy programming for decentralized two-level linear
programming problems. Fuzzy Sets Syst 125:301–315
Sakawa M, Nishizaki I (2002b) Interactive fuzzy programming for two-level nonconvex programming
problems with fuzzy parameters through genetic algorithms. Fuzzy Sets Syst 127:185–197
226 M. Sakawa et al.
123
Sakawa M, Nishizaki I (2009) Cooperative and noncooperative multi-level programming. Springer, New
York
Sakawa M, Yano, H (1985) Interactive fuzzy satisficing method using augumented minimax problems
and its application to environmental systems. IEEE Trans Syst Man Cybern SMC 15:720–729
Sakawa M, Yano H (1989) Interactive decision making for multiobjective nonlinear programming
problems with fuzzy parameters. Fuzzy Sets Syst 29:315–326
Sakawa M, Yano H (1990) An interactive fuzzy satisficing method for generalized multiobjective linear
programming problems with fuzzy parameters. Fuzzy Sets Syst 35:125–142
Sakawa M, Yano H, Yumine T (1987) An interactive fuzzy satisficing method for multiobjective linear-
programming problems and its application. IEEE Trans Syst Man Cybern SMC 17:654–661
Sakawa M, Nishizaki I, Uemura Y (1998) Interactive fuzzy programming for multi-level linear
programming problems. Comput Math Appl 36:71–86
Sakawa M, Nishizaki I, Uemura Y (2000) Interactive fuzzy programming for multi-level linear
programming problems with fuzzy parameters. Fuzzy Sets Syst 109:3–19
Sakawa M, Nishizaki I, Uemura Y (2001) Interactive fuzzy programming for two-level linear and linear
fractional production and assignment problems: a case study. Eur J Oper Res 135:142–157
Sakawa M, Nishizaki I, Uemura Y (2002) A decentralized two-level transportation problem in a housing
material manufacturer—interactive fuzzy programming approach. Eur J Oper Res 141:167–185
Sakawa M, Kato K, Nishizaki I (2003) An interactive fuzzy satisficing method for multiobjective
stochastic linear programming problems through an expectation model. Eur J Oper Res
145:665–672
Sakawa M, Kato K, Katagiri H (2004) An interactive fuzzy satisficing method for multiobjective linear
programming problems with random variable coefficients through a probability maximization
model. Fuzzy Sets Syst 146:205–220
Sakawa M, Kato K, Azad MAK, Watanabe R (2005) A genetic algorithm with double string for nonlinear
integer programming problems. In: IEEE SMC 2005 conference proceedings. pp 3281–3286
Shih HS, Lai YJ, Lee ES (1996) Fuzzy approach for multi-level programming problems. Comput Oper
Res 23:73–91
Shimizu K, Ishizuka Y, Bard JF (1997) Nondifferentiable and two-level mathematical programming.
Kluwer, Boston
Sinha S (2003) Fuzzy programming approach to multi-level programming problems. Fuzzy Sets Syst
136:189–202
Slowinski R (ed) (1998) Fuzzy sets in decision analysis, operations research and statistics. Kluwer,
Dordrecht/Boston/London
Slowinski R, Teghem J (eds) (1990) Stochastic versus fuzzy approaches to multiobjective mathematical
programming under uncertainty. Kulwer, Dordrecht/Boston/London
Stancu-Minasian IM (1990) Overview of different approaches for solving stochastic programming
problems with multiple objective functions. In: Slowinski R, Teghem J (eds) Stochastic versus fuzzy
approaches to multiobjective mathematical programming under uncertainty, Kluwer, Dordrecht/
Boston/London, pp 71–101
Zimmermann H-J (1978) Fuzzy programming and linear programming with several objective functions.
Fuzzy Sets Syst 1:45–55
Zimmermann H-J (1987) Fuzzy sets, decision-making and expert systems. Kluwer, Boston
Interactive fuzzy stochastic two-level integer programming 227
123