International Journal of Fuzzy Systems, Vol. 12, No.

1, March 2010 15

Fuzzy Generalized Hybrid Aggregation Operators and its Application in

Fuzzy Decision Making
José M. Merigó and Montserrat Casanovas

Abstract1 OWA operator by using generalized means. Recently

[17], a further generalization has been suggested by us-
The hybrid averaging (HA) is an aggregation op- ing order inducing variables in the GOWA operator. This
erator that uses the weighted average (WA) and the operator is known as the induced GOWA (IGOWA) op-
ordered weighted averaging (OWA) operator in the erator. The main advantage of this operator is that it
same formulation. In this paper, we introduce several generalizes a wide range of mean operators such as the
generalizations of the HA operator by using general- induced OWA (IOWA), the induced ordered weighted
ized and quasi-arithmetic means, fuzzy numbers and quadratic averaging (IOWQA) and the generalized mean
order inducing variables in the reordering step of the (GM). The IGOWA operator can be further generalized
aggregation process. We present the fuzzy general- by using quasi-arithmetic means [17]. The result is the
ized hybrid averaging (FGHA) operator, the fuzzy Quasi-IOWA operator. Other generalizations of the OWA
induced generalized hybrid averaging (FIGHA) op- operator by using generalized or quasi-arithmetic means
erator, the Quasi-FHA operator and the Quasi-FIHA can be found in [3-4,8,12,14,16-18,38].
operator. The main advantage of these operators is Going a step further, in this paper we introduce the
that they generalize a wide range of fuzzy aggrega- fuzzy generalized hybrid averaging (FGHA) and the
tion operators that can be used in a wide range of ap- fuzzy induced generalized hybrid averaging (FIGHA)
plications such as decision making problems. For operator. They are new aggregation operators that use
example, we could mention the fuzzy induced hybrid the main characteristics of the IGOWA, the FOWA and
averaging (FIHA), the fuzzy weighted generalized the HA operator. Thus, these operators use generalized
mean (FWGM) and the fuzzy induced generalized means in the HA operator and in uncertain situations
OWA (FIGOWA). We end the paper with an applica- where the available information can not be represented
tion of the new approach in a decision making prob- with exact numbers but it is possible to use FNs. By us-
lem. ing the HA operator, the FIGHA considers the WA and
the IOWA (or the OWA) in the same problem. In deci-
Keywords: Aggregation operators, Fuzzy numbers, sion making problems, this implies that the FIGHA op-
Hybrid averaging, OWA operator, Decision making. erator considers the subjective probability and the attitu-
dinal character of the decision maker in the same formu-
1. Introduction lation. One of the key features of the FGHA and the
FIGHA operator is that they generalize a wide range of
The hybrid averaging (HA) operator [1] is an aggrega- aggregation operators such as the FGM, the FWGM, the
tion operator that uses the weighted average (WA) and FIGOWA, the FWA, the FOWA, the fuzzy induced HA
the ordered weighted averaging (OWA) operator [2-38] (FIHA), the fuzzy induced hybrid quadratic averaging
in the same formulation. An interesting extension of the (FIHQA), the fuzzy induced hybrid geometric averaging
HA operator for situations where it is not possible to use (FIHGA) and the fuzzy induced hybrid harmonic aver-
exact numbers because the information is uncertain, is aging (FIHHA) operator. Thus, they are able to provide a
the fuzzy HA (FHA) operator. It represents the uncertain more complete view of the decision problem to the deci-
information to be aggregated with the HA operator by sion maker, so he will be able to make the appropriate
using FNs. decision according to his interests.
Another interesting aggregation operator is the gener- We further generalize the FGHA and the FIGHA by
alized OWA (GOWA) operator [8, 29]. It generalizes the using quasi-arithmetic means. The result is the
Quasi-FHA and the Quasi-FIHA operator. We also de-
Corresponding Author: José M. Merigó is with the Department of velop a numerical example of the new approach in a de-
Business Administration, University of Barcelona, Av. Diagonal 690, cision making problem about selection of strategies.
08034 Barcelona, Spain. With this example, we will see that depending on the
E-mail: [email protected]
Manuscript received 13 Aug. 2008; revised 20 Oct. 2009; accepted 24 aggregation operator used, the results may lead to dif-
Feb. 2010. ferent decisions.

© 2010 TFSA
16 International Journal of Fuzzy Systems, Vol. 12, No. 1, March 2010

In order to do so, this paper is organized as follows. In It can be defined as follows.

Section 2 we briefly describe some basic concepts about Definition 1: Let Ψ be the set of FNs. A FHA operator of
the FNs, the FHA and the IGOWA operator. In Section 3 dimension n is a mapping FHA: Ψn → Ψ that has an as-
we present the FGHA and the FIGHA operator. Section 4 sociated weighting vector W of dimension n such that the
analyses some basic families of the FIGHA operator. sum of the weights is 1 and wj ∈ [0, 1], then:
Section 5 introduces the Quasi-FHA and the Quasi-FIHA n
operators. In Section 6 we develop a numerical example FHA (ã1, ã2, …, ãn)= ∑ w j b j (2)
of the new approach. Finally, in Section 7 we summarize j =1
the main conclusions of the paper. where bj is the jth largest of the âi (âi = nωiãi, i =
1,2,…,n), ω = (ω1, ω2, …, ωn)T is the weighting vector of
2. Preliminaries the ãi, with ωi ∈ [0, 1] and the sum of the weights is 1,
and the ãi are FNs.
In this Section, we give a brief overview of the FNs, Note that the FHA operator can be extended by using
the FHA and the IGOWA operator. order inducing variables, obtaining the fuzzy induced
HA (FIHA) operator. It can be defined as follows.
A. Fuzzy Numbers Definition 2: Let Ψ be the set of FNs. A FIHA operator
The FN was first introduced by [39-40]. Since then, it of dimension n is a mapping FIHA: Ψn × Ψn → Ψ that
has been studied and applied by a lot of authors such as has an associated weighting vector W of dimension n
[41-42]. with the sum of the weights equal to 1 and wj ∈ [0, 1],
A FN is a fuzzy subset of a universe of discourse that such that:
is both convex and normal. Note that the FN may be n
considered as a generalization of the interval number FIHA (〈u1,ã1〉, 〈u2,ã2〉, …, 〈un,ãn〉)= ∑ w j b j (3)
although it is not strictly the same because the interval j =1
numbers may have different meanings. In the literature, where bj is the âi value (âi = nωiãi, i = 1,2,…,n) of the
we find a wide range of FNs [41-45]. For example, a IOWA pair 〈ui, ãi〉 having the jth largest ui, ui is the order
trapezoidal FN (TpFN) A of a universe of discourse R inducing variable, ω = (ω1, ω2, …, ωn)T is the weighting
can be characterized by a trapezoidal membership func- vector of the ãi, with ωi ∈ [0, 1] and the sum of the
tion A = (a, a ) such that weights is 1, and the ãi are FNs.
a(α ) = a1 + α (a 2 − a1 ), Similar to the FOWA operator, it is possible to analyze
(1) different properties of the FHA operator. Note that in this
a (α ) = a 4 − α (a 4 − a 3 ).
case, we should also consider the problem of comparing
where α ∈ [0, 1] and parameterized by (a1, a2, a3, a4) FNs in the reordering process. For simplicity, we rec-
where a1 ≤ a2 ≤ a3 ≤ a4, are real values. Note that if a1 = ommend to follow the policy explained in [12-13].
a2 = a3 = a4, then, the FN is a crisp value and if a2 = a3,
the FN is represented by a triangular FN (TFN). Note C. The Induced Generalized OWA Operator
that the TFN can be parameterized by (a1, a2, a4). The induced generalized OWA (IGOWA) operator [17]
In the following, we are going to review some basic is a generalization of the OWA operator by using gener-
FN arithmetic operations as follows. Let A and B be two alized means and order inducing variables. It includes a
TFN, where A = (a1, a2, a3) and B = (b1, b2, b3). Then: wide range of mean operators such as the usual OWA,
1) A + B = (a1 + b1, a2 + b2, a3 + b3). the IOWA, the induced OWG operator, the ordered
2) A − B = (a1 − b3, a2 − b2, a3 − b1). weighted quadratic averaging operator (OWQA), etc. It
3) A × k = (k × a1, k × a2, k × a3); for k > 0. can be defined as follows.
Note that other operations could be studied [39-45] Definition 3: An IGOWA operator of dimension n is a
but in this paper we will focus on these ones. mapping IGOWA: Rn × Rn → R defined by an associated
weighting vector W of dimension n such that
B. Fuzzy Hybrid Averaging Operator
The FHA operator is an aggregation operator that uses ∑ nj =1 w j = 1 and wj ∈ [0, 1], a set of order-inducing
the weighted average (WA) and the OWA operator in the variables ui, and a parameter λ ∈ (−∞, ∞), according to
same formulation. Then, it is possible to consider in the the following formula:
same decision making problem, the attitudinal character 1/ λ
⎛ n ⎞
of the decision maker and its subjective probability. It IGOWA (〈u1,a1〉, …, 〈un,an〉)= ⎜ ∑ w j bλj ⎟ (4)
also deals with uncertain environments where the avail- ⎜ j =1 ⎟
⎝ ⎠
able information cannot be assessed with exact numbers where (b1, …, bn) is (a1, a2,…, an) reordered in decreasing
but it is possible to find approximate results using FNs. order of the values of the ui, the ui are the order-inducing
José M. Merigó et al.: Fuzzy Generalized Hybrid Aggregation Operators and its Application in Fuzzy Decision Making 17

variables, ai are the argument variables. complex reordering processes in the aggregation of the
For further reading on generalized OWAs and its re- FNs. It uses order-inducing variables in the reordering of
cent developments, see for example [3-4,8,12,14,16- the FNs and it includes the FGHA as a particular case. It
18,38]. can be defined as follows.
If we replace bλ with a general continuous strictly Definition 6: Let Ψ be the set of FNs. A FIGHA operator
monotone function g(b) [3], then, the IGOWA operator of dimension n is a mapping FIGHA: Ψn × Ψn → Ψ that
becomes the Quasi-IOWA operator [17]. It can be for- has an associated weighting vector W of dimension n
mulated as follows. with ∑ nj =1 w j = 1 and wj ∈ [0, 1], such that:
Definition 4: A Quasi-IOWA operator of dimension n is a
1/ λ
mapping QIOWA: Rn × Rn → R defined by an associated ⎛ n ⎞
weighting vector W of dimension n such that FIGHA (〈u1,ã1〉, …, 〈un,ãn〉)= ⎜ ∑ w j b λj ⎟ (7)
⎜ j =1 ⎟
⎝ ⎠
∑nj=1 w j = 1 and wj ∈ [0, 1], and by a strictly monotonic
where bj is the âi value (âi = nωiãi, i = 1,2,…,n) of the
continuous function g(b), as follows: IOWA pair 〈ui, ãi〉 having the jth largest ui, ui is the order
⎛ n ⎞ inducing variable, ω = (ω1, ω2, …, ωn)T is the weighting
⎜ j =1
( )
QIOWA (〈u1,a1〉, …, 〈un,an〉)= g −1 ⎜ ∑ w j g b( j ) ⎟ (5)
⎟ vector of the ãi, with ωi ∈ [0, 1] and the sum of the
⎝ ⎠
weights is 1, the ãi are FNs, and λ is a parameter such
where the bj are the argument values ai of the
that λ ∈ (−∞, ∞).
Quasi-IOWA pairs 〈ui, ai〉 ordered in decreasing order of
As we can see the FGHA operator is a particular case
their ui, values.
of the FIGHA operator when the reordering of the ui is
equal to j.
3. Fuzzy Generalized Hybrid Aggregation Op- Note that if λ ≤ 0, we can only use positive numbers
erators R+, in order to get consistent results. Note also that dif-
ferent types of FNs could be used in the aggregation
In this Section, we are going to consider two gener- such as Triangular FNs, Trapezoidal FNs, L-R FNs, in-
alizations: the fuzzy generalized hybrid averaging terval-valued FNs, intuitionistic FNs [21,39], general-
(FGHA) operator and the fuzzy induced generalized hy- ized FNs, and more complex structures
brid averaging (FIGHA) operator. As the FIGHA gener- [12-13,19,24-25,31,39-45].
alizes the FGHA, we will mostly use the FIGHA opera- In this case, the reordering of the arguments has an
tor in this paper. The FGHA operator is an extension of additional difficulty because now we are using FNs.
the HA operator that uses generalized means and uncer- Therefore, in some cases it is not clear which FN is
tain information that can be represented by using FNs. higher, so we need to establish an additional criteria for
Thus, we can represent the problems considering the best ranking FNs. For simplicity, we recommend the follow-
and worst result and the possibility that the internal val- ing procedure. Select the FN with the highest value in
ues of the FN will occur. With this generalization, we the membership level α = 1. Note that if the membership
include in the same formulation a lot of aggregation op- level α = 1 is an interval, then, we will calculate the av-
erators such as the FHA, the fuzzy hybrid quadratic av- erage of the interval. Note also that this problem is espe-
eraging (FHQA), the FGOWA and the FWGM. It can be cially relevant for the FGHA because the FIGHA solves
defined as follows. this issue with the order inducing variables that do not
Definition 5: Let Ψ be the set of FNs. A FGHA operator need to rank FNs. However, the final result is a FN so if
of dimension n is a mapping FGHA: Ψn → Ψ that has an we have different final results, then, we need to establish
associated weighting vector W of dimension n such that a ranking between them.
the sum of the weights is 1 and wj ∈ [0, 1], then: From a generalized perspective of the reordering step,
1/ λ it is possible to distinguish between the descending
⎛ n ⎞
FGHA (ã1, ã2, …, ãn)= ⎜ ∑ w j bλj ⎟ (6) FIGHA (DFIGHA) and the ascending FIGHA (AFIGHA)
⎜ j =1 ⎟
⎝ ⎠ operator. The weights of these operators are related by wj
where bj is the jth largest of the âi (âi = nωiãi, i = = w*n−j+1, where wj is the jth weight of the DFIGHA and
1,2,…,n), ω = (ω1, ω2, …, ωn)T is the weighting vector of w*n−j+1 the jth weight of the AFIGHA operator.
the ãi, with ωi ∈ [0, 1] and the sum of the weights is 1, If B is a vector corresponding to the ordered argu-
the ãi are FNs and λ is a parameter such that λ ∈ (−∞, ments bjλ, we shall call this the ordered argument vector
T
∞). and W is the transpose of the weighting vector, then, the
The FIGHA operator is an extension of the FGHA op- FIGHA operator can be expressed as:
erator for situations where we want to deal with more
18 International Journal of Fuzzy Systems, Vol. 12, No. 1, March 2010

FIGHA (〈u1,ã1〉, …, 〈un,ãn〉)= W T B ( )1 / λ

(8)
these measures that takes into account both weighting
vectors.
Note that if the weighting vector is not normalized,
i.e., W = ∑nj=1 w j ≠ 1 , then, the FIGHA operator can be 4. Families of FIGHA Operators
expressed as:
1/ λ
In this Section, we analyze different families of
⎛ 1 n ⎞ FIGHA operators. We will distinguish between those
FIGHA (〈u1,ã1〉, …, 〈un,ãn〉)= ⎜ ∑ w j b λj ⎟ (9) found in the weighting vector W and those found in the
⎜ W j =1 ⎟
⎝ ⎠ parameter λ. The main advantage of using these families
The FIGHA operator is a mean or averaging operator. is that they can be very useful for the decision maker in
This is a reflection of the fact that the operator is mono- some specific situations. However, each family is just
tonic and idempotent. These properties can be proved one particular case. Therefore, they can only be used in
with the following theorems. some particular situations but they cannot be seen as a
Theorem 1 (Monotonicity): Assume f is the FIGHA op- general model that can be used in different frameworks
erator, if ãi ≥ ei, for all ãi, then: such as the general formulation explained in the previous
f (〈u1,ã1〉, …, 〈un,ãn〉) ≥ f (〈u1,e1〉, …, 〈un,en〉) (10) Section. Note that these families could also be studied
Proof: Let from the perspective of the FGHA operator.
1/ λ
⎛ n ⎞
f (〈u1,ã1〉, …, 〈un,ãn〉)= ⎜ ∑ w j b λj ⎟ (11) A. Analyzing the Weighting Vector W
⎜ j =1 ⎟
⎝ ⎠ By using a different weighting vector in the FIGHA
1/ λ operator, we are able to obtain different types of fuzzy
⎛ n ⎞
f (〈u1,e1〉, …, 〈un,en〉)= ⎜ ∑ w j d λj ⎟ (12) aggregation operators. For example, it is possible to ob-
⎜ j =1 ⎟ tain the fuzzy hybrid maximum, the fuzzy hybrid mini-
⎝ ⎠
Since ãi ≥ ei, for all ãi, it follows that, ãi ≥ ei, and then mum, the fuzzy generalized mean (FGM), the fuzzy
f (〈u1,ã1〉, …, 〈un,ãn〉) ≥ f (〈u1,e1〉, …, 〈un,en〉) ■ weighted generalized mean (FWGM), the FIGOWA op-
Theorem 2 (Idempotency): Assume f is the FIGHA op- erator and the FGHA operator.
erator, if ãi = ã, for all ãi, then: Remark 1: The fuzzy hybrid maximum is obtained if wp
f (〈u1,ã1〉, …, 〈un,ãn〉) = ã (13) = 1 and wj = 0, for all j ≠ p, and up = Max{ãi}. The fuzzy
Prof: Since ãi = ã, for all ãi, we have hybrid minimum is obtained if wp = 1 and wj = 0, for all j
1/ λ ≠ p, and up = Min{ai}. More generally, if wk = 1 and wj
⎛ n ⎞ = 0, for all j ≠ k, we get for any λ, FIGHA (〈u1,ã1〉,
f (〈u1,ã1〉, 〈u2,ã2〉…, 〈un,ãn〉)= ⎜ ∑ w j b λj ⎟ =
⎜ j =1 ⎟ 〈u2,ã2〉…, 〈un,ãn〉) = bk, where bk is the kth largest argu-
⎝ ⎠
1/ λ 1/ λ ment ãi. The FGM is found when wj = 1/n, and ωi = 1/n,
⎛ n ⎞ ⎛ n ⎞ for all ãi. The FWGM is obtained when wj = 1/n, for all
= ⎜ ∑ w j aλ ⎟ = ⎜ aλ ∑ w j ⎟ (14)
⎜ j =1 ⎟ ⎜ j =1 ⎟ ãi. The FIGOWA is found when ωi = 1/n, for all ãi. The
⎝ ⎠ ⎝ ⎠
FGHA operator is formed when the reordering of the
n
Since ∑ j =1 w j = 1, we get order inducing variables ui is equal to bj. That is, to the
f (〈u1,ã1〉, 〈u2,ã2〉, …, 〈un,ãn〉)=ã ■ reordering of the jth largest argument.
Note that the FGHA and the FIGHA operator are not Remark 2: Following a similar methodology as it has
commutative because the weighted average (a part of the been developed in [2-4,9,11-17,23,26-30,35], we could
hybrid aggregation) is not commutative. They are not study other particular cases of the FIGHA operator. For
strictly bounded because in some particular situations example, when wj* = 1/m for k ≤ j* ≤ k + m − 1 and wj* =
[12], the aggregation may be lower than the minimum 0 for j* > k + m and j* < k, we are using the win-
and higher than the maximum. dow-FIGHA operator. Note that k and m must be posi-
A further interesting issue to consider are the meas- tive integers such that k + m − 1 ≤ n.
ures for characterizing the weighting vector W of the Remark 3: The olympic-FIGHA is found when w1 = wn =
FGHA and the FIGHA operator such as the attitudinal 0, and for all others wj* = 1/(n − 2). A general form of the
character, the entropy of dispersion, the divergence of W olympic-FIGHA can be used considering that wj = 0 for j
and the balance operator. Note that these measures fol- = 1, 2, …, k, n, n − 1, …, n − k + 1; and for all others wj*
low the same methodology as the original version de- = 1/(n − 2k), where k < n/2. Note that if k = 1, then,
veloped for the OWA and the GOWA operator [12,26,29] this general form becomes the usual olympic aggregation.
but they do not affect directly the weighting vector ω. If k = (n − 1)/2, then, this general form becomes the me-
Note also that it is possible to use an hybrid version of dian-FIGHA aggregation. That is, if n is odd we assign
José M. Merigó et al.: Fuzzy Generalized Hybrid Aggregation Operators and its Application in Fuzzy Decision Making 19

w(n + 1)/2 = 1 and wj* = 0 for all others. If n is even we as- harmonic averaging (FIHHA) operator and the fuzzy
sign for example, wn/2 = w(n/2) + 1 = 0.5 and wj* = 0 for all induced hybrid quadratic averaging (FIHQA) operator.
others. Remark 9: When λ = 1, we get the FIHA operator.
Remark 4: Another type of aggregation that could be n
used is the E-Z FIGHA weights. In this case, we should FIGHA (〈u1,ã1〉, 〈u2,ã2〉, …, 〈un,ãn〉)= ∑ w j b j (15)
j =1
distinguish between two classes. In the first class, we
assign wj* = (1/q) for j* = 1 to q and wj* = 0 for j* > q, From a generalized perspective of the reordering step
and in the second class, we assign wj* = 0 for j* = 1 to n we can distinguish between the DFIHA operator and the
− q and wj* = (1/q) for j* = n − q + 1 to n. If q = 1 for the AFIHA operator. Note that if wj = 1/n, for all ãi, we get
first class, the E-Z FIGHA becomes the fuzzy hybrid the FWA and if ωj = 1/n, for all ãi, we get the FIOWA
maximum. And if q = 1 for the second class, the E-Z operator. If wj = 1/n, and ωj = 1/n, for all ãi, then, we get
FIGHA becomes the fuzzy hybrid minimum. the fuzzy average (FA).
Remark 5: For the weighted median-FIGHA, we select Remark 10: When λ = 0, the FIGHA operator becomes
the argument bk that has the kth largest argument such the FIHGA operator.
that the sum of the weights from 1 to k is equal or higher n
FIGHA (〈u1,ã1〉, 〈u2,ã2〉, …, 〈un,ãn〉)= ∏ b j w j (16)
than 0.5 and the sum of the weights from 1 to k − 1 is j =1
less than 0.5. In this case, it is also possible to distinguish between
Remark 6: A further interesting family is the S-FIGHA descending (DFIHGA) and ascending (AFIHGA) orders.
operator based on the S-OWA operator [27]. It can be Note that if wj = 1/n, for all ãi, we get the fuzzy weighted
subdivided in three classes: the “orlike”, the “andlike” geometric average (FWGA) and if ωj = 1/n, for all ãi, we
and the generalized S-FIGHA operator. The generalized get the fuzzy induced OWG (FIOWG) operator. If wj =
S-FIGHA operator is obtained when w1 = (1/n)(1 − (α + 1/n, and ωj = 1/n, for all ãi, then, we get the fuzzy geo-
β)) + α, wn = (1/n)(1 − (α + β)) + β, and wj = (1/n)(1 − metric mean (FGM).
(α + β)) for j = 2 to n − 1 where α, β ∈ [0, 1] and α + β Remark 11: When λ = −1, we get the FIHHA operator.
≤ 1. Note that if α = 0, the generalized S-FIGHA opera- 1
tor becomes the “andlike” S-FIGHA operator and if β = FIGHA (〈u1,ã1〉, 〈u2,ã2〉, …, 〈un,ãn〉)= (17)
n wj
0, it becomes the “orlike” S-FIGHA operator. ∑
Remark 7: Another family of aggregation operator that j =1 b j
could be used is the centered-FIGHA operator. We could Note that we can distinguish between the descending
define a FIGHA operator as a centered aggregation op- FIHHA (DFIHHA) operator and the ascending FIHHA
erator if it is symmetric, strongly decaying and inclusive. (AFIHHA) operator. Note that if wj = 1/n, for all ãi, we
Note that these properties have to be accomplished for get the fuzzy weighted harmonic mean (FWHM) and if
the weighting vector w of the FIOWA operator but not ωj = 1/n, for all ãi, we get the fuzzy induced ordered
necessarily for the weighting vector ω of the FWA. It is weighted harmonic averaging (FIOWHA) operator. If wj
symmetric if wj = wj+n−1. It is strongly decaying when i < = 1/n, and ωj = 1/n, for all ãi, then, we get the fuzzy
j ≤ (n + 1)/2 then wi < wj and when i > j ≥ (n + 1)/2 then harmonic mean (FHM).
wi < wj. It is inclusive if wj > 0. Note that it is possible to Remark 12: When λ = 2, we get the FIHQA operator.
consider a softening of the second condition by using wi 1/ 2
⎛ n ⎞
≤ wj instead of wi < wj. We shall refer to this as softly FIGHA (〈u1,ã1〉, …, 〈un,ãn〉) = ⎜ ∑ w j b 2j ⎟ (18)
decaying centered-FIGHA operator. Another particular ⎜ j =1 ⎟
⎝ ⎠
situation of the centered-FIGHA operator appears if we In this case, we also get the descending FIHQA
remove the third condition. We shall refer to it as a (DFIHQA) operator and the ascending FIHQA (AFI-
non-inclusive centered-FIGHA operator. HQA) operator. If wj = 1/n, for all ãi, we get the fuzzy
Remark 8: Other families of FIGHA operators could be weighted quadratic average (FWQA) and if ωj = 1/n, for
studied such as the Gaussian FIGHA weights, the non- all ãi, we get the fuzzy induced OWQA (FIOWQA) op-
monotonic FIGHA operator, etc. For more information, erator. If wj = 1/n, and ωj = 1/n, for all ãi, then, we get
see for example [2-4,9,11-17,23,26-30,35]. the fuzzy quadratic average (FQA).
Remark 13: Note that other families could be obtained
B. Analyzing the Parameter λ by using different values in the parameter λ. Note also
If we analyze different values of the parameter λ, we that it is possible to study these families individually.
obtain another group of particular cases such as the usual Then, we could develop for each case, a similar analysis
FIHA operator, the fuzzy induced hybrid geometric av- as it has been developed in Section 3 and 4.1, where we
eraging (FIHGA) operator, the fuzzy induced hybrid study different properties and families of the fuzzy hy-
20 International Journal of Fuzzy Systems, Vol. 12, No. 1, March 2010

brid aggregation operators. i.e., W = ∑nj=1 w j ≠ 1 , then, the Quasi-FIHA operator

5. The Quasi-FHA and the Quasi-FIHA Opera- can be expressed as:

tors Quasi-FIHA (〈u1,ã1〉, …, 〈un,ãn〉)=
⎛1 n ⎞
As it is explained in [3], it is possible to further gener-
g −1 ⎜ ∑
⎜ W j =1
( )
w j g b( j ) ⎟
⎟
(21)
alize the GOWA operator by using quasi-arithmetic ⎝ ⎠
means. Following the same methodology as in [3,6,17], Note that all the properties and particular cases com-
we can suggest a similar generalization to the FGHA and mented in the FIGHA operator, are also included in these
the FIGHA operators by using quasi-arithmetic means. generalizations. Thus, for example, we could mention
We will call these generalizations the Quasi-FHA and the the problem of reordering the arguments when they are
Quasi-FIHA operator. The Quasi-FHA is defined as fol- FNs. In order to solve this problem, we recommend the
lows. method explained in [12-13].
Definition 7: Let Ψ be the set of FN. A Quasi-FHA op-
erator of dimension n is a mapping QFHA: Ψn → Ψ that 6. Numerical Example
has an associated weighting vector W of dimension n
In the following, we are going to develop a numerical
such that the sum of the weights is 1 and wj ∈ [0, 1],
example in order to illustrate the new approach. We will
then:
consider a decision making problem where a company is
⎛ n ⎞
⎜ j =1
( )
Quasi-FHA(ã1, ã2…, ãn)= g −1 ⎜ ∑ w j g b( j ) ⎟ (19)
⎟
analyzing the optimal strategy for the next period. We
will use different types of FIGHA (or Quasi-FIHA) op-
⎝ ⎠
erators such as the FA, the FWA, the FOWA, the FHA
where bj is the jth largest of the âi (âi = nωiãi, i =
and the FIHA operator.
1,2,…,n), ω = (ω1, ω2, …, ωn)T is the weighting vector of Assume that a company that it is operating in Europe
the ãi, with ωi ∈ [0, 1] and the sum of the weights is 1, and North America is analyzing its general policy for the
the ãi are FNs and g(b) is a strictly continuous mono- next year and they consider 5 possible strategies to fol-
tonic function. low.
The Quasi-FIHA operator extends the Quasi-FHA op- • A1: Expand to the Asian market.
erator by using order inducing variables. Thus, we get a
• A2: Expand to the South American market.
more complete fuzzy generalized aggregation operator. It
• A3: Expand to the African market.
can be defined as follows.
• A4: Do not develop any expansion.
Definition 8: Let Ψ be the set of FN. A Quasi-FIHA op-
• A5: Expand to the 3 continents.
erator of dimension n is a mapping QFIHA: Ψn × Ψn →
In order to evaluate these strategies, the group of ex-
Ψ that has an associated weighting vector W of dimen-
perts of the company considers that the key factor for the
sion n such that ∑ nj =1 w j = 1 and wj ∈ [0, 1], then: next year is the economic situation. Then, depending on
the situation, the expected benefits for the company will
Quasi-FIHA (〈u1,ã1〉, 〈u2,ã2〉…, 〈un,ãn〉)=
be different. The experts have considered 5 possible
⎛ n ⎞
⎜ j =1
( )
g −1 ⎜ ∑ w j g b( j ) ⎟
⎟
(20) situations for the next year: S1 = Very bad, S2 = Bad, S3 =
Normal, S4 = Good, S5 = Very good. The results are
⎝ ⎠
shown in Table 1. Note that the results are TFNs.
where bj is the âi value (âi = nωiãi, i = 1,2,…,n) of the
IOWA pair 〈ui, ãi〉 having the jth largest ui, ui is the order Table 1. Fuzzy payoff matrix.
inducing variable, ω = (ω1, ω2, …, ωn)T is the weighting
vector of the ãi, with ωi ∈ [0, 1] and the sum of the A1
S1
(20,30,40)
S2
(40,50,60)
S3
(30,40,50)
S4
(50,60,70)
S5
(60,70,80)
weights is 1, the ãi are FNs, and g(b) is a strictly con- A2 (60,70,80) (30,40,50) (50,60,70) (50,60,70) (20,30,40)
tinuous monotonic function. A3 (40,50,60) (60,70,80) (20,30,40) (50,60,70) (30,40,50)
(60,70,80) (60,70,80) (50,60,70) (30,40,50) (10,20,30)
As we can see, we replace bλ of the FIGHA with a
A4
A5 (30,40,50) (60,70,80) (60,70,80) (30,40,50) (30,40,50)
general continuous strictly monotone function g(b). In
this case, the weights of the ascending and descending In this example, we assume that the experts use the
versions are also related by wj = w*n−j+1, where wj is the following weighting vector for all the cases of the WA
jth weight of the Quasi-DFIHA and w*n−j+1 the jth and the OWA operators: W = (0.1, 0.2, 0.2, 0.2, 0.3).
weight of the Quasi-AFIHA operator. Note that they assume the following order inducing
Note that if the weighting vector is not normalized, variables: U = (14, 20, 40, 33, 28).
With this information, it is possible to aggregate it in
José M. Merigó et al.: Fuzzy Generalized Hybrid Aggregation Operators and its Application in Fuzzy Decision Making 21

order to take a decision. The results are presented in Ta- Table 4. Fuzzy aggregated results using the FOWA operator.
ble 2. α A1 A2 A3 A4 A5
0 (36, 56) (38, 58) (36, 56) (37, 57) (39, 59)
Table 2. Fuzzy aggregated results. 0.1 (37, 55) (39, 57) (37, 55) (38, 56) (40, 58)
0.2 (38, 54) (40, 56) (38, 54) (39, 55) (41, 57)
FA FWA FOWA FHA FIHA 0.3 (39, 53) (41, 55) (39, 53) (40, 54) (42, 56)
A1 (40,50,60) (44,54,64) (36,46,56) (36,45,54) (42,52,62) 0.4 (40, 52) (42, 54) (40, 52) (41, 53) (43, 55)
A2 (42,52,62) (38,48,58) (38,48,58) (36,45.5,55) (36,46,56) 0.5 (41, 51) (43, 53) (41, 51) (42, 52) (44, 54)
A3 (40,50,60) (39,49,59) (36,46,56) (35,44.5,54) (39,49,59) 0.6 (42, 50) (44, 52) (42, 50) (43, 51) (45, 53)
A4 (42,52,62) (37,47,57) (37,47,57) (32.5,43,53) (35,45,55) 0.7 (43, 49) (45, 51) (43, 49) (44, 50) (46, 52)
A5 (42,52,62) (42,52,62) (39,49,59) (37.5,47,56.5) (37.5,47.5,57.5) 0.8 (44, 48) (46, 50) (44, 48) (45, 49) (47, 51)
0.9 (45, 47) (47, 49) (45, 47) (46, 48) (48, 50)
1 46 48 46 47 49
Note that these results can also be expressed using the
α-cut representation that represents the membership Table 5. Fuzzy aggregated results using the FIHA operator.
function of the FN as it has been explained in Section
α
2.1. Recall that for the TFN A = (a, a ) , we use the fol- 0
A1
(42, 62)
A2
(36, 56)
A3
(39, 59)
A4
(35, 55)
A5
(37.5, 57.5)
lowing equations: 0.1 (43, 61) (37, 55) (40, 58) (36, 54) (38.5, 56.5)
0.2 (44, 60) (38, 54) (41, 57) (37, 53) (39.5, 55.5)
a(α ) = a1 + α (a 2 − a1 ), 0.3 (45, 59) (39, 53) (42, 56) (38, 52) (40.5, 54.5)
(22)
a (α ) = a3 − α (a3 − a 2 ). 0.4 (46, 58) (40, 52) (43, 55) (39, 51) (41.5, 53.5)
0.5 (47, 57) (41, 51) (44, 54) (40, 50) (42.5, 52.5)
where α ∈ [0, 1] and parameterized by (a1, a2, a3) where 0.6 (48, 56) (42, 50) (45, 53) (41, 49) (43.5, 51.5)
0.7 (49, 55) (43, 49) (46, 52) (42, 48) (44.5, 50.5)
a1 ≤ a2 ≤ a3, are real values. 0.8 (50, 54) (44, 48) (47, 51) (43, 47) (45.5, 49.5)
In order to present the results in a more complete way, 0.9 (51, 53) (45, 47) (48, 50) (44, 46) (46.5, 48.5)
we calculate the intervals obtained in the α-cut repre- 1 52 46 49 45 47.5

sentation for each decimal between [0, 1]. Therefore, we

are able to analyze the results obtained for different de- Table 6. Ordering of the strategies.
grees of membership. These results are presented in Ta- Ordering
bles 3, 4 and 5. Note that we focus on the results ob- FA A2=A4=A5⎬A1=A3
FWA A1⎬A5⎬A3⎬A2⎬A4
tained in the FWA, the FOWA and the FIHA operator. FOWA A5⎬A2⎬A4⎬A1=A3
As we can see in Tables 2, 3, 4 and 5, depending on FHA A5⎬A2⎬A1⎬A3⎬A4
the aggregation operator used, the optimal strategy may FIHA A1⎬A3⎬A5⎬A2⎬A4
be different.
A further interesting issue is to establish an ordering As we can see, we get different orderings depending
of the strategies. This is very useful when we want to on the aggregation operator used. The main advantage of
consider more than one alternative. Note that if the or- this analysis is that the decision maker gets a more com-
dering is not clear due to the FNs, we will have to estab- plete view of the different scenarios that could happen in
lish a criterion for comparing FNs. As it has been ex- the future depending on the method used. Although he
plained in Section 3, we recommend following the will select the alternative that it is in accordance with his
method explained in [12-13] where we calculate an av- interests, he will be concerned on other potential results
erage or a weighted average of the FNs. The results are that could happen in the uncertain environment.
shown in Table 6.
7. Conclusions
Table 3. Fuzzy aggregated results using the FWA operator.
α A1 A2 A3 A4 A5 We have presented different generalizations of the hy-
0 (44, 64) (38, 58) (39, 59) (37, 57) (42, 62) brid aggregation operators such as the FGHA and the
0.1 (45, 63) (39, 57) (40, 58) (38, 56) (43, 61) FIGHA operators. We have seen that they are extensions
0.2 (46, 62) (40, 56) (41, 57) (39, 55) (44, 60)
0.3 (47, 61) (41, 55) (42, 56) (40, 54) (45, 59) of the HA operator that use generalized means and un-
0.4 (48, 60) (42, 54) (43, 55) (41, 53) (46, 58) certain information assessed with FNs. Moreover, we
0.5 (49, 59) (43, 53) (44, 54) (42, 52) (47, 57)
0.6 (50, 58) (44, 52) (45, 53) (43, 51) (48, 56)
have seen that they include the WA and the OWA opera-
0.7 (51, 57) (45, 51) (46, 52) (44, 50) (49, 55) tor (and the IOWA) in the same formulation in a similar
0.8 (52, 56) (46, 50) (47, 51) (45, 49) (50, 54) way than the HA operator. Particularly, these operators
0.9 (53, 55) (47, 49) (48, 50) (46, 48) (51, 53)
1 54 48 49 47 52 include the FWGM and the FIGOWA operator by using
the hybrid formulation. It also includes a lot of other
types of aggregation operators such as the FHA, the
FHGA, the FHQA, the FGOWA and the FGM.
We have further generalized this approach by using
22 International Journal of Fuzzy Systems, Vol. 12, No. 1, March 2010

quasi-arithmetic means. As a result, we have obtained operators,” International Journal of Approximate

the Quasi-FHA and the Quasi-FIHA operators. We have Reasoning, vol. 45, no. 1, pp. 68-81, 2007.
also developed an application of these generalized ag- [10] X. Liu, “Parameterized defuzzification with con-
gregation operators in a decision making problem con- tinuous weighted quasi-arithmetic means: An ex-
cerning the selection of strategies. Note that a lot of tension,” Information Sciences, vol. 179, no. 8, pp.
other applications could be developed with the FGHA 1193-1206, 2009.
and the FIGHA such as in statistics, operational research, [11] X. Liu and S. Han, “Orness and parameterized RIM
engineering and economics. quantifier aggregation with OWA operators: A
summary,” International Journal of Approximate
Acknowledgment Reasoning, vol. 48, no. 1, pp. 77-97, 2008.
[12] J. M. Merigó, New extensions to the OWA opera-
We would like to thank the Editor-in-Chief, the Asso- tors and its application in decision making, PhD
ciate Editor and the anonymous referees for their valu- thesis (In Spanish), Department of Business Ad-
able comments that have improved the quality of the ministration, University of Barcelona, 2008.
paper. Support from the project MC238206 is gratefully [13] J. M. Merigó and M. Casanovas, “Using fuzzy
acknowledged. numbers in heavy aggregation operators,” Interna-
tional Journal of Information Technology, vol. 4,
References no. 3, pp. 177-182, 2008.
[14] J. M. Merigó and M. Casanovas, “The induced
[1] Z. S. Xu and Q. L. Da, “An overview of operators generalized hybrid averaging operator and its ap-
for aggregating information,” International Journal plication in decision making,” International Jour-
of Intelligent Systems, vol. 18, no. 9, pp. 953-969, nal of Business, Economics, Finance and Man-
2003. agement Science, vol. 1, no. 2, pp. 95-101, 2009.
[2] B. S. Ahn, “Preference relation approach for ob- [15] J. M. Merigó and M. Casanovas, “Induced aggre-
taining OWA operator weights,” International gation operators in decision making with Demp-
Journal of Approximate Reasoning, vol. 47, no. 2, ster-Shafer belief structure,” International Journal
pp. 166-178, 2008. of Intelligent Systems, vol. 24, no. 8, pp. 934-954,
[3] G. Beliakov, “Learning weights in the generalized 2009.
OWA operators,” Fuzzy Optimization and Decision [16] J. M. Merigó and A.M. Gil-Lafuente, “The gener-
Making, vol. 4, no.2, pp. 119-130, 2005. alized adequacy coefficient and its application in
[4] G. Beliakov, A. Pradera, and T. Calvo, Aggregation strategic decision making,” Fuzzy Economic Re-
Functions: A guide for practitioners, Berlin: view , vol. 13, no. 2, pp. 17-36, 2008.
Springer-Verlag, 2007. [17] J. M. Merigó and A.M. Gil-Lafuente, “The induced
[5] H. Dyckhoff and W. Pedrycz, “Generalized means generalized OWA operator,” Information Sciences,
as a model of compensative connectives,” Fuzzy vol. 179, no. 6, pp. 729-741, 2009.
Sets and Systems, vol. 14, no. 2, pp. 143-154, 1984. [18] J. H. Wang and J. Hao, “A new version of 2-tuple
[6] J. Fodor, J. L. Marichal, and M. Roubens, “Char- fuzzy linguistic representation model for comput-
acterization of the ordered weighted averaging op- ing with words,” IEEE Transactions on Fuzzy Sys-
erators,” IEEE Transactions on Fuzzy Systems, vol. tems , vol. 14, no. 3, pp. 435-445, 2006.
3, no. 2, pp. 236-240, 1995. [19] X. Wang, “Fuzzy number intuitionistic fuzzy
[7] Y. C. Hu and J. F. Tsai, “Fusing fuzzy association arithmetic aggregation operators,” International
rule-based classifiers using Sugeno integrals with Journal of Fuzzy Systems, vol. 10, no. 2, pp.
ordered weighted averaging operators,” Interna- 104-111, 2008.
tional Journal of Uncertainty, Fuzziness and [20] Y. M. Wang and Y. Luo, “Generalised fuzzy
Knowledge-Based Systems, vol. 15, no. 6, pp. weighted mean and its application,” International
717-735, 2007. Journal of General Systems, vol. 38, no. 5, pp.
[8] N. Karayiannis, “Soft learning vector quantization 533-546, 2009.
and clustering algorithms based on ordered [21] G. W. Wei, “Uncertain linguistic hybrid mean op-
weighted aggregation operators,” IEEE Transac- erator and its application to group decision making
tions on Neural Networks, vol. 11, no. 5, pp. under uncertain linguistic environment,” Interna-
1093-1105, 2000. tional Journal of Uncertainty, Fuzziness and
[9] X. Liu, “The solution equivalence of minimax dis- Knowledge-Based Systems, vol. 17, no. 2, pp.
parity and minimum variance problems for OWA 251-267, 2009.
José M. Merigó et al.: Fuzzy Generalized Hybrid Aggregation Operators and its Application in Fuzzy Decision Making 23

[22] J. Wu, J. C. Cai, H. Li, and W. Q. Duan, “The in- [37] M. Zarghami, F. Szidarovszky, and R. Ardakanian,
duced continuous ordered weighted geometric op- “A fuzzy-stochastic OWA model for robust
erators and their application in group decision multi-criteria decision making,” Fuzzy Optimiza-
making,” Computers & Industrial Engineering, vol. tion and Decision Making, vol. 7, no. 1, pp. 11-15,
56, no. 4, pp. 1545-1552, 2009. 2008.
[23] Z. S. Xu, “Dependent uncertain ordered weighted [38] H. Zhao, Z. S. Xu, M. F. Ni, and S. Liu, “General-
aggregation operators,” Information Fusion, vol. 9, ized aggregation operators for intuitionistic fuzzy
no. 3, pp. 310-316, 2008. sets,” International Journal of Intelligent Systems,
[24] Z. S. Xu, “Fuzzy harmonic mean operators,” In- vol. 25, no. 1, pp. 1-30, 2010.
ternational Journal of Intelligent Systems, vol. 24, [39] S. S. L. Chang and L.A. Zadeh, “On fuzzy map-
no. 2, pp. 152-172, 2009. ping and control,” IEEE Transactions on Systems,
[25] Z. S. Xu and R.R. Yager, “Dynamic intuitionistic Man and Cybernetics, vol. 2, no. 1, pp. 30-34,
fuzzy multi-attribute decision making,” Interna- 1972.
tional Journal of Approximate Reasoning, vol. 48, [40] L. A. Zadeh, “The concept of a linguistic variable
no. 1, pp. 246-262, 2008. and its application to approximate reasoning,” Part
[26] R. R. Yager, “On ordered weighted averaging ag- 1, Information Sciences, vol. 8, no. 2, pp. 199-249,
gregation operators in multi-criteria decision mak- 1975; Part 2, Information Sciences, vol. 8, no. 4, pp.
ing,” IEEE Transactions on Systems, Man and Cy- 301-357, 1975; Part 3, Information Sciences, vol. 9,
bernetics B, vol. 18, no.1, pp. 183-190, 1988. no. 1, pp. 43-80, 1975.
[27] R. R. Yager, “Families of OWA operators,” Fuzzy [41] D. Dubois and H. Prade, Fuzzy Sets and Systems:
Sets and Systems, vol. 59, no. 2, pp. 125-148, 1993. Theory and Applications, New York: Academic
[28] R. R. Yager, “On weighted median aggregation,” Press, 1980.
International Journal of Uncertainty, Fuzziness [42] A. Kaufmann and M. M. Gupta, Introduction to
and Knowledge-Based Systems, vol. 2, no. 1, pp. fuzzy arithmetic, Rheinhold: Publications Van
101-113, 1994. Nostrand, 1985.
[29] R. R. Yager, “Generalized OWA aggregation op- [43] T. Y. Chen and T. C. Ku, “Importance-assessing
erators,” Fuzzy Optimization and Decision Making, method with fuzzy number-valued fuzzy measures
vol. 3, no. 1, pp. 93-107, 2004. and discussions on TFNs and TrFNs,” International
[30] R. R. Yager, “Centered OWA operators,” Soft Journal of Fuzzy Systems, vol. 10, no. 2, pp.
Computing, vol. 11, no. 7, pp. 631-639, 2007. 92-103, 2008.
[31] R. R. Yager, “Time series smoothing and OWA ag- [44] Z. Chiang and G.H. Tzeng, “A third party logistics
gregation,” IEEE Transactions on Fuzzy Systems, provider for the best selection in fuzzy dynamic
vol. 16, no. 4, pp. 994-1007, 2008. decision environments,” International Journal of
[32] R. R. Yager, “Using trapezoids for representing Fuzzy Systems, vol. 11, no. 1, pp. 1-9, 2009.
granular objects: Applications to learning and OWA [45] Y. M. Wang and K. S. Chin, “A linear goal pro-
aggregation,” Information Sciences, vol. 178, no. 2, gramming priority method for fuzzy analytic hier-
pp. 363-380, 2008. archy process and its applications in new product
[33] R. R. Yager, “On generalized Bonferroni mean op- screening,” International Journal of Approximate
erators for multi-criteria aggregation,” Interna- Reasoning, vol. 49, no. 3, pp. 451-465, 2008.
tional Journal of Approximate Reasoning, vol. 50,
no. 8, pp. 1279-1286, 2009. José M. Merigó has a MSc and a PhD
[34] R. R. Yager and D.P. Filev, “Induced ordered degree in Business Administration ob-
weighted averaging operators,” IEEE Transactions tained in 2003 and 2009, respectively,
on Systems, Man and Cybernetics B, vol. 29, no. 1, from University of Barcelona, Spain. He
also holds a Bachelor Degree in Eco-
pp. 141-150, 1999.
nomics obtained in 2005, from Lund
[35] R. R. Yager, and J. Kacprzyk, The Ordered University, Sweden. He is an Assistant
Weighted Averaging Operators: Theory and Appli- Professor in the Department of Business
cations, Norwell: Kluwer Academic Publishers, Administration at the University of Bar-
1997. celona. He has published more than 100 papers in journals and
[36] D. Y. Yeh, C. H. Cheng, and H. W. Yio, “Empirical conference proceedings including journals such as Information
research of the principal component analysis and Sciences, International Journal of Intelligent Systems and
ordered weighted averaging integrated evaluation Computers & Industrial Engineering. He is currently inter-
model on software projects,” Cybernetics and Sys- ested in Aggregation Operators, Decision Making and Uncer-
tems, vol. 38, no. 3, pp. 289-303, 2007. tainty.
24 International Journal of Fuzzy Systems, Vol. 12, No. 1, March 2010

Montserrat Casanovas has a MSc and a

PhD degree in Business Administration
obtained in 1969 and 1977, respectively.
The MSc was obtained at the University
of Barcelona and the PhD at the
Autonomous University of Barcelona.
She is also a European Financial Adviser
and a Full Professor in the Department of
Business Administration at the Univer-
sity of Barcelona. She has published more than 100 papers in
journals and conference proceedings including journals such
as International Journal of Intelligent Systems, Fuzzy Eco-
nomic Review and the Journal of International Business Dis-
ciplines. She is currently interested in Financial Management,
Decision Making and Uncertainty.