1. Introduction
Ranking fuzzy numbers constitutes a well studied research topic due to its significant applications in approximate reasoning [1], decision-making [2], forecasting [3], and other fuzzy systems. As a fundamental problem within fuzzy decision-making, it has drawn considerable research interest. Numerous ranking methodologies have been proposed, and early comprehensive reviews are available in [4]-[6]. Representative works include the relative-preference-relation-based approach for triangular and trapezoidal fuzzy numbers developed by Wang [7], as well as the possibility-theory-driven method reported by Gu and Xuan [8].
As summarized by Ma and Yang [9], existing fuzzy number ranking techniques can be grouped into three classes: based on defuzzification methods, based on ranking-index methods, and pairwise-comparison methods. Although based on defuzzification methods are computationally convenient, heuristic manipulation may lead to the loss of inherent fuzzy information [10]. In comparison, based on ranking-index and pairwise-comparison schemes preserve full fuzzy information yet suffer from higher computational complexity. Along this line, Yuan [11] advocated constructing fuzzy ranking via fuzzy preference relations. Wang [7] further pointed out that such preference relations are capable of quantifying preference degrees with membership functions.
Preference assessment for fuzzy numbers is closely linked to the ranking of interval-valued quantities, which has also received substantial research attention. Interval numbers serve as a powerful tool for modelling expert judgements [12], and interval-valued weights are widely adopted in various decision-making scenarios [13]. Moreover, interval ordering occupies a critical position in multiple-attribute decision-making [14].
This paper focuses on constructing order relations for fuzzy numbers. Inspired by the properties of interval total orders, several partial orders for fuzzy numbers are established, and their fundamental characteristics are investigated. To address the incomparability limitation inherent to partial orders, we further develop a set of total order relations specifically for trapezoidal fuzzy numbers. These total orders enable clear discrimination of priority levels among distinct fuzzy numbers. Finally, numerical examples concerning fuzzy number ranking are presented to validate the feasibility and effectiveness of the proposed total order relations. The findings offer theoretical underpinnings for real-world applications such as multi-attribute fuzzy decision-making and fuzzy games.
2. Preliminaries
Proposed by Zadeh [15], a fuzzy set defined on a nonempty set
is a mapping
. For any element
, the value
is termed the membership degree of
, which characterizes the degree to which
belongs to the fuzzy set
. As a special and widely investigated class of fuzzy sets on the real number field
, fuzzy numbers extend the conventional definition of real numbers. Different from precise single-valued real numbers, fuzzy numbers permit a set of plausible values to describe uncertain quantities. The fundamental concept of fuzzy numbers was first proposed and systematically investigated by Dubois and Prade [16].
Definition 1. ([16]) A fuzzy set
on
is called a fuzzy number, if it satisfies the following conditions:
1)
is normal, i.e., there exists
such that
;
2)
is fuzzy convex, i.e.,
for all
and
;
3)
is upper semi-continuous on
, i.e.,
,
with
whenever
; and
4) the closure of
, i.e.,
is bounded. Where the symbol
is a topological closure operator with respect to the usual topology on
.
Throughout this paper,
denotes the collection of all fuzzy numbers. For any fuzzy number
and each
, the
-level set of
is defined as
. The kernel of
, denoted by
, is given by
, which is equivalent to the 1-level set
. In addition, the 0-level set
corresponds to the support
of the fuzzy number
. It is well known that a fuzzy number can be uniquely determined by its family of
-level sets. Accordingly, for any two fuzzy numbers
,
holds if and only if
for all
.
Let
represent the set of all nonempty, closed, and bounded intervals, which is formally defined as:
Each interval
can be correspondingly regarded as a specific fuzzy number via its characteristic function. Specifically, the fuzzy valued extension
of the interval
is defined as:
Due to its rectangular membership distribution,
is termed a rectangular fuzzy number. In the degenerate case where
, the single-point interval
degenerates to a standard real number. For simplicity, we adopt the notation
instead of
or
for such degenerate intervals. In this framework, closed real intervals can be regarded as a special class of fuzzy numbers, such that
. This further implies the embedding relationship
, demonstrating that fuzzy numbers serve as a generalized form of real numbers and real intervals. According to Definition 1, the
-level set
of any fuzzy number
is a nonempty closed interval for all
. This interval is uniformly written as
, where the finite real numbers
and
satisfy
, and are defined as the left and right endpoints of
, respectively. Xiao and Wang [17] systematically summarized the fundamental properties of the
-level sets (
) of fuzzy numbers and established an equivalent characterization principle for fuzzy numbers based on these interval families, which is presented in the following proposition.
Proposition 1 ([17]) A fuzzy set
on
is a fuzzy number if and only if
is a non-empty and closed interval for every
.
A trapezoidal fuzzy number is defined as a quadruple
, where
and
are all real numbers satisfying
, and its membership function is given by
Its
-level set takes the form
In the special case where
and
,
will degenerate as a closed interval and is denoted by
or
. Throughout this paper,
denotes the collection of all trapezoidal fuzzy numbers.
3. Some Total Orders of Trapezoidal Fuzzy Numbers
Let
be a nonempty set. A binary relation
defined on
is termed a partial order if it satisfies reflexivity, antisymmetry and transitivity.
is referred to as an ordered set (or partially ordered set). Note that an order
on
gives rise to a relation
of strict inequality:
on
if and only if
and
. Furthermore, an order
in
is called total or linear if, for any
,
or
.
3.1. Some Total Orders of Closed and Bounded Intervals
For two intervals
, if we compare their left and right endpoints, then the following component-wise order (also named
order in [18]) is defined.
Definition 2.
if and only if
and
,
if and only if
and
.
On the set
, the component-wise order is a frequently used partial order. Some pairs of intervals remain incomparable, which complicates the identification of a unique optimal interval solution in practical problems. To mitigate this difficulty, some total orders were introduced for the set of closed subintervals of
that refines the component-wise order. Some important classes of admissible orders in
are defined as following.
1) Lexicographical orders.
(1)
(2)
2) Xu-Yager order with preference for higher uncertainty. If the decision maker prefers the alternative with the larger expected value, and when expectations coincide, the one with greater uncertainty, the
order [19] is
(3)
3) Xu-Yager order with preference for lower uncertainty. Conversely, if the decision maker still favors the larger expected value but when expectations coincide, prefers lower uncertainty, the order is
(4)
In Figure 1(a), the collection of bounded closed intervals, composed of both the orange and blue ones, represents the set of intervals that are smaller than
according to the order
. Specifically, these intervals satisfy the ranking criteria defined by
relative to
. In Figure 1(b), the set of orange and blue intervals corresponds to the collection of those bounded closed intervals that are ranked lower than
under the order
. This order
determines the relative size of these intervals with respect to
. For Figure 1(c), the set of orange and blue bounded closed intervals precisely constitutes the set of all intervals that are smaller than
in terms of the order
. That is, based on the rules of the
order, these intervals are considered to be of a smaller magnitude compared to
.
![]()
Figure 1. Blue region is the set of intervals smaller than
under the order
; Orange and blue regions in (a), (b) and (c) are the set of intervals smaller than
under the order
,
and
respectively.
3.2. Some Orders of Fuzzy Numbers and Trapezoidal Fuzzy Numbers
For two fuzzy numbers
, Klir and Yuan [20] introduced the KY order:
which can equivalently be expressed via the component-wise order
on
-level sets:
(5)
For two TrFNs
, from Definitions 1, 2 and Equation (5), we have
(6)
Figure 2(a) shows a case where
, the meaning expressed by the
order is that
is greater than
, Figure 2(b) shows a case where
.
Figure 2. The red graph and the blue graph represent membership functions of trapezoidal fuzzy numbers
and
, respectively. (a)
; (b)
,
are not comparable in
and
,
.
As previously stated, establishing a total order for fuzzy numbers is quite intricate. In the subsequent part of this section, we will construct total orders specifically for
. In Equations (1-4), we know
,
,
and
are total orders in partially ordered sets
. We can generalize these four orders from
to the set of fuzzy numbers as follows.
1) Compare the left endpoints of each
-level set. If the left endpoints are equal, compare their right endpoints. The lexicographical orders are defined as follows:
,
(7)
(8)
2) If experts prioritize alternatives with a higher expected value for each
-level set, and in case of equal expected values, prefer those with greater uncertainty, the order
is defined as:
,
(9)
3) If experts prioritize alternatives with a higher expected value for each
-level set, and in case of equal expected values, prefer those with lower uncertainty, the order
is defined as:
,
(10)
Theorem 1.
,
,
and
are partial orders on fuzzy numbers, moreover
,
,
and
whenever
.
Proof. We shall only prove that
is a partial order and
whenever
, with the proofs for the remaining three following analogously. Let
,
,
be fuzzy numbers.
1)
(reflexivity) is trivial.
2) Suppose
and
, then
, 1)
, or
and 2)
, or
hold. Thus,
. Therefore,
(antisymmetry) holds.
3) Suppose
and
, then
, a)
, or
and b)
, or
hold. Thus,
, or
. Therefore,
(transitivity) holds.
Suppose
, then
,
. Therefore,
. Thus,
, or
holds, i.e.,
. □
The above theorem proves that Equations. (7-10) define a partial order, yet they do not constitute a total order. Figure 2(b) shows a case where
,
are not comparable in
and
,
.
If
and
are trapezoidal fuzzy numbers, we can define
,
,
and
as follows.
(11)
(12)
(13)
(14)
Theorem 2.
,
,
and
are total orders trapezoidal fuzzy numbers, moreover
whenever
,
whenever
,
whenever
and
whenever
in
.
Proof. We shall only prove that
is a total order and
whenever
, with the proofs for the remaining three following analogously. Let
,
and
be trapezoidal fuzzy numbers.
1)
(reflexivity) is trivial.
2) Suppose
and
, then from Equation (11), we have
and
,
. Thus,
(antisymmetry) holds.
3) Suppose
and
, then from Equation (11), we have
and
,
. Thus,
. a) If
or
, then
. Thus,
. b) If
, and
, then
. Thus,
. c) If
, and
, then
. Thus,
. (d) If
, and
, then
. Thus,
. By a similar argument, the conclusion is valid for the other six cases. Therefore,
(transitivity) holds.
Suppose
, then
,
, or
. Therefore,
, or
. Thus,
, or
, or
, or
holds, i.e.,
. □
4. Application to Rank Trapezoidal Fuzzy Numbers
In this section, we apply the main results derived in the aforementioned sections to the ranking of trapezoidal fuzzy numbers. A total order imposes clear, well-defined sequential relationships on the elements of a set. In a totally ordered set, any two elements can be compared in terms of their sequence with absolute clarity, i.e., leaving no room for ambiguity or uncertainty regarding their order. Consequently, the findings of this paper can be applied to the ranking of trapezoidal fuzzy numbers.
The four trapezoidal fuzzy numbers featured in the subsequent example are drawn from [21]. This example is primarily intended to demonstrate that these total orders can be used for ranking trapezoidal fuzzy numbers. Furthermore, this approach is able to resolve issues that remain unaddressed by the magnitude-based methods proposed in [22] [23]. Notably, when evaluated from different perspectives, the total orders in the set of trapezoidal fuzzy numbers lead to distinct conclusions regarding the ranking of these four trapezoidal fuzzy numbers.
Example 1. Consider the four trapezoidal fuzzy numbers
,
,
and
, which are graphically represented in Figure 3, where
is green, and
is red, and
is blue, and
is yellow.
Figure 3. Membership functions of the trapezoidal fuzzy numbers
,
,
and
.
If we aim to ascertain which of these four trapezoidal fuzzy numbers is larger with respect to the magnitude relationship among the numbers, we can utilize the
order for the comparison. Nevertheless, the ultimate result of this comparison shows that, within the framework of the
order, any two of these four trapezoidal fuzzy numbers are incomparable. Further, the following results can be obtained.
1)
and
, but
and
are incomparable in the order
. And
.
2)
and
, but
and
are incomparable in the order
. And
.
Example 2. This example taken from ([9] [10]). The collection
of fuzzy sets with triangular membership functions. The parameter values of
are shown in Table 1.
Table 1. Trapezoidal fuzzy numbers in Example 2.
A1 |
(33, 65, 68, 74) |
A6 |
(65, 68, 72, 83) |
A2 |
(55, 58, 62, 73) |
A7 |
(22, 80, 83, 86) |
A3 |
(67, 76, 125, 130) |
A8 |
(28, 31, 34, 61) |
A4 |
(38, 67, 83, 107) |
A9 |
(2, 5, 20, 66) |
A5 |
(72, 85, 88, 91) |
A10 |
(11, 70, 73, 76) |
1) Based on a parametric relation pair, the method proposed by Dombi and Jonas [10] enables the ranking of a collection of trapezoidal fuzzy numbers. Applying this method, the resulting ranking is determined as:
. Here, the notation
indicates that
and
are non-comparable.
2) Ma et al. [9] put forward a novel approach for ranking fuzzy numbers, which is built on additively consistent fuzzy preference relations. When this method is adopted, the specific ranking of the fuzzy numbers is obtained as:
.
3) To judge the ordering relationship among these ten trapezoidal fuzzy numbers according to their magnitude relations, the
order should be adopted for comparison. As mentioned above, since the
order is a partial order, incomparable cases may exist. Therefore, the total order defined by Equations (11-14) can be used for comparison. If Equation (13) is employed, the resulting ranking is determined as:
.
5. Conclusion
In this paper, multiple partial order relations for fuzzy numbers are constructed based on the characteristics of interval total orders. Aiming at the incomparability defect of traditional partial order relations, a corresponding total order relation system for trapezoidal fuzzy numbers is further established in this study. This system effectively solves the problem of priority indistinguishability between fuzzy numbers and realizes the ordered discrimination of different fuzzy numbers. Numerical examples of fuzzy number ranking verify the feasibility and practical effectiveness of the proposed total order relations. The research results can provide important theoretical basis and technical support for the practical applications of fuzzy systems such as multi-attribute fuzzy decision-making and fuzzy games.
Future research can examine the performance of the proposed ranking orders under standard arithmetic operations (addition and scalar multiplication) to validate the robustness of the ranking function. It is also valuable to explore practical implications of inconsistent ranking outcomes and develop guidance for interpreting gaps between the proposed approach and benchmark methods in real decision scenarios. Moreover, the current framework is limited to a single geometric fuzzy set type; extending it to widely adopted continuous fuzzy representations like bell curves and Gaussian distributions merits further investigation. Additionally, future theoretical work may adapt the strict hierarchical parameter-checking scheme to nonlinear and more complex fuzzy models.