IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 16, NO.

5, OCTOBER 2008 1379

Short Papers
Modeling Random Fuzzy Renewal Reward Processes interarrival times. This includes an extended version of the renewal
equation, Blackwell’s renewal theorem, and Smith’s key renewal theo-
Qiang Shen, Ruiqing Zhao, and Wansheng Tang rem, all in fuzzy random form. Li et al. [10] introduced the concept of
a fuzzy random delayed renewal process in which the interarrival times
between two events are characterized as fuzzy random variables. They
Abstract—This short paper discusses the modeling of random fuzzy
renewal reward processes in which the interarrival times and rewards are
also defined a fuzzy random equilibrium renewal process as a special
represented by nonnegative random fuzzy variables. Based on random case of the fuzzy random delayed renewal process [10]. By dealing with
fuzzy theory, a random fuzzy variable denotes a measurable function from interarrival times as exponentially distributed fuzzy random variables,
a credibility space to a collection of random variables. Under this setting, Li et al. [9] considered a fuzzy random homogeneous Poisson process
the long-run expected reward per unit time is addressed and the theorem and a fuzzy random compound Poisson process, and established several
on random fuzzy renewal rewards is established. The utility of this research
is demonstrated with a realistic application case. important properties of these two types of processes.
The other approach is to use the random fuzzy theory developed
Index Terms—Credibility measure, fuzzy variables, interarrival time, by Liu [12]. Briefly, a random fuzzy variable is a measurable function
random fuzzy variables, renewal processes, stochastic processes.
from a possibility space to a collection of random variables. In this
approach, the expected value operator of random fuzzy variables was
I. INTRODUCTION introduced in [18]. Based on this theory, Shen et al. [24] conducted
an investigation into the modeling of alternating renewal processes.
Renewal reward processes play an important role in process theory.
This resulted in a theorem on the limit value of the average chance of
A large number of models, such as dispatching a train and queuing
a given random fuzzy event, in terms of “a system being on at time
in a bank, are special cases of these processes. Thus, modeling such
t.” These renewal processes were represented by sequences of positive
processes has a significant value in real-world applications. Having
random fuzzy vectors. Zhao et al. [31] considered a random fuzzy
recognized this, stochastic renewal reward processes have been well
process in which the interarrival times were modeled as i.i.d. random
developed [22]. Indeed, based on probability theory, much research has
fuzzy variables, and described certain properties of the random fuzzy
been reported to perform such modeling (e.g., [1], [2], [5], [22], [26],
renewal processes, including the random fuzzy elementary renewal
and [27]). In representing stochastic renewal processes, an underlying
theorem and Blackwell’s theorem. As a continuation of the significant
assumption is that the interarrival times and rewards are deemed to
initial work of [31], this short paper considers the modeling of another
be random variables. However, in many cases, consideration of ran-
important type of random fuzzy process—the random fuzzy renewal
domness alone will not help to evaluate a process in a satisfactory
reward process.
manner. This is because fuzziness and randomness in one process are
The short paper is organized as follows. For completeness, Section II
often mixed up, and it is not easy to distinguish between them from a
briefly outlines the background of fuzzy variables and random fuzzy
practical viewpoint. They are usually required to be considered simul-
variables within the framework of the credibility theory [12], in terms
taneously.
of basic concepts and properties relevant to the present short paper.
There are two approaches to deal with these kinds of phenomena.
Section III describes the random fuzzy renewal reward processes and
One is the fuzzy random theory first introduced by Kwakernaak [7], [8].
presents the random fuzzy renewal reward theorem. Section IV shows a
Roughly speaking, a fuzzy random variable is a mathematical descrip-
worked example, illustrating the effectiveness of the proposed theorem.
tor for a fuzzy random phenomenon and defined as a measurable func-
Section V concludes the short paper.
tion from a probability space to a collection of fuzzy sets. Based on
fuzzy random theory, the modeling of typical fuzzy random processes
has been considered by several authors. Hwang [6] investigated a re- II. FUZZY VARIABLES AND RANDOM FUZZY VARIABLES
newal process in which the interarrival times were considered indepen- To cope with uncertain and imprecise process modeling, the pos-
dent and identically distributed (i.i.d.) fuzzy random variables and a sibility measure [28] or the necessity measure [4], [23] is commonly
theorem on the fuzzy rate of fuzzy random renewal processes was pro- used. In fact, the main difference between these two measures is that
vided. Popova and Wu [21] considered a renewal reward process with they consider the same question from different angles. The possibility
fuzzy random interarrival times and rewards. Also, the long-run aver- measure assesses a given vague concept in terms of affirmation, and the
age fuzzy reward per unit time was stated in [21]. Zhao and Tang [30] necessity measure does so in terms of disaffirmation. However, their
addressed some important properties of fuzzy random renewal pro- use may lead to situations where the former overrates the possibility
cesses, which are generated by a sequence of i.i.d. fuzzy random for a vague concept to be correctly captured while the latter underrates
such a possibility. To have a balanced approach, and based on the basic
intuitions behind the possibility and necessity measures, a self-dual
Manuscript received April 7, 2008; accepted July 12, 2008. First published
August 26, 2008; current version published October 8, 2008. This work was sup- measure, called credibility, has been introduced [16], [17].
ported in part by the U.K. Engineering and Physical Sciences Research Council Given a universe Θ, let P(Θ) denote the power set of Θ. The
(EPSRC) under Grant EP/D057086 and in part by the National Natural Science credibility measure of a set A, Cr{A}, can be defined [16] such that
Foundation of China under Grant 70571056, Grant 70471049, and the Program
for New Century Excellent Talents in University. 1
Q. Shen and R. Zhao are with the Department of Computer Science, Aberys- Cr{A} = (Pos{A} + Nec{A}) (1)
twyth University, Aberystwyth SY23 3DB, U.K. (e-mail: [email protected]; 2
[email protected]).
W. Tang is with the Institute of Systems Engineering, Tianjin University,
where Pos and Nec represent the possibility and necessity measures,
Tianjin 300072, China (e-mail: [email protected]). respectively. The triplet (Θ, P(Θ), Cr) is called a credibility space. The
Digital Object Identifier 10.1109/TFUZZ.2008.2005014 rest of this section introduces the notions and notations closely relevant

1063-6706/$25.00 © 2008 IEEE

1380 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 16, NO. 5, OCTOBER 2008

to the development of the work to be presented in the next section, based Let (Ω, A, Pr) be a probability space and F be a collection of
on the concepts of credibility space and credibility measure. random variables defined on probability space (Ω, A, Pr). Then the
Definition 1: A fuzzy variable is defined as a function from the following definitions can be introduced and propositions established.
credibility space (Θ, P(Θ), Cr) to the real line
. Definition 6 ([12]): A random fuzzy variable is a measurable func-
Remark 1 ([13]): Let ξ be a fuzzy variable on the credibility space tion from a credibility space (Θ, P(Θ), Cr) to a collection of random
(Θ, P(Θ), Cr). Then its membership function can be derived from the variables F.
credibility measure such that Example 1: Consider an example of a random variable as given
in [12]. Let ξ be the lifetime of a system. Usually, the probability
µ(x) = min(2Cr{ξ = x}, 1), x ∈
. (2) distribution of ξ is assumed to be known completely except for the
values of certain parameters. For instance, it might be known that
Definition 2 ([12]): Let ξ be a fuzzy variable and α ∈ (0, 1]. Then
the lifetime ξ is an exponentially distributed random variable with an
ξα = inf{x|µ(x) ≥ α} and ξα = sup{x|µ(x) ≥ α} (3) unknown mean λ
1
e−x / λ , if 0 ≤ x < ∞
are called the α-pessimistic value and the α-optimistic value of ξ, φ(x) = λ
respectively. 0, otherwise.
Proposition 1 ([18]): Let ξ and η be two fuzzy variables. Then:
In statistics, an interval estimate or point estimate of the value of λ is
1) for any α ∈ (0, 1], (ξ + η)α = ξα + ηα ;
provided by experimental data. In many practical situations, however,
2) for any α ∈ (0, 1], (ξ + η)α = ξα + ηα .
such data are often unavailable. If the value of λ is represented as a
Furthermore, if ξ and η are nonnegative (i.e., Cr{ξ < 0} = 0 and
fuzzy variable, defined on the credibility space (Θ, P(Θ), Cr), then ξ
Cr{η < 0} = 0), then:
is a random fuzzy variable that can be described by
3) for any α ∈ (0, 1], (ξ · η)α = ξα · ηα ;
4) for any α ∈ (0, 1], (ξ · η)α = ξα · ηα . ξ(λ(θ)) ∼ EX P(λ(θ)) (8)
Definition 3 ([16], [17]): Let ξ be a fuzzy variable. The expected
value E[ξ] of ξ is defined as where λ is a fuzzy variable on (Θ, P(Θ), Cr) and the symbol EX P is
an abbreviation of the exponential distribution.

∞
0 Proposition 5 ([15]): Let ξ be a random fuzzy variable defined on
E[ξ] = Cr{ξ ≥ r}dr − Cr{ξ ≤ r}dr (4) the credibility space (Θ, P(Θ), Cr). Then, we have the following.
−∞
0
1) The probability Pr{ξ(θ) ∈ B} is a fuzzy variable for any Borel
provided that at least one of the two integrals is finite (avoiding the set B of
.
case ∞ − ∞). Especially, if ξ is a nonnegative fuzzy variable, then 2) The expected value E[ξ(θ)] is a fuzzy variable provided that
∞
E[ξ] = 0 Cr{ξ ≥ r}dr. E[ξ(θ)] is finite for each θ ∈ Θ.
Proposition 2 ([18]): Let ξ be a fuzzy variable with finite expected Definition 7: A random fuzzy variable ξ defined on the cred-
value E[ξ]. Then ibility space (Θ, P(Θ), Cr) is said to be positive if and only if

Pr{ξ(θ) ≤ 0} = 0 for each θ ∈ Θ with Cr{θ} > 0.
1
1 Definition 8 ([18]): Let ξ be a random fuzzy variable defined on the
E[ξ] = [ξα + ξα ] dα (5)
2 0
credibility space (Θ, P(Θ), Cr). The expected value E[ξ] is defined
by
where ξα and ξα are the α-pessimistic value and the α-optimistic value
∞
of ξ, respectively. E[ξ] = Cr{θ ∈ Θ | E[ξ(θ)] ≥ r}dr
Definition 4 ([19]): The fuzzy variables ξ1 , ξ2 , . . . , ξn are said to be 0
independent if
0
− Cr{θ ∈ Θ | E[ξ(θ)] ≤ r}dr (9)
Cr{ξi ∈ Bi , i = 1, 2, . . . , n} = inf Cr{ξi ∈ Bi } (6) −∞
1 ≤i ≤n
provided that at least one of the two integrals is finite.Especially, if ξ
∞
for any sets B1 , B2 , . . . , Bn of
. is a nonnegative random fuzzy variable, then E[ξ] = 0 Cr{θ ∈ Θ |
Definition 5 ([15]): The fuzzy variables ξ1 , ξ2 , . . . , ξn are said to be E[ξ(θ)] ≥ r}dr.
identically distributed if Remark 2 ([12]): If the random fuzzy variable ξ degenerates to a
random variable, then the expected value operator becomes
Cr{ξi ∈ B} = Cr{ξj ∈ B}, i, j = 1, 2, . . . , n (7)


∞ 0

for any set B of
. E[ξ] = Pr{ξ ≥ r}dr − Pr{ξ ≤ r}dr (10)

−∞
Proposition 3 ([14]): Suppose (Θi , P(Θi ), Cri ), i =1, 2, . . . , n, 0
n
are credibility spaces. Let Θ = Θ1 × Θ2 × · · · × Θn = i = 1 Θi and which is just the conventional mathematical expectation of random
Cr{A} = sup(θ 1 , θ 2 , . . . , θ n )∈A inf i Cri {θi }, A ∈ P(Θ). Then the set variable ξ. If the random fuzzy variable ξ degenerates to a fuzzy vari-
function Cr is a credibility measure on P(Θ) and (Θ, P(Θ), Cr) able, then the expected value operator becomes
is a credibility space (called the product credibility space of
∞
0
(Θi , P(Θi ), Cri ), i = 1, 2, . . . , n). E[ξ] = Cr{ξ ≥ r}dr − Cr{ξ ≤ r}dr (11)
Proposition 4 ([15]): Let (Θi , P(Θi ), Cri ), i =1, 2, . . . , be an 0 −∞
∞
arbitrary sequence of credibility spaces and Θ = i = 1 Θi . Define which is just the expected value of fuzzy variable ξ.
Cr on P(Θ) such that Cr{A} = sup(θ 1 , θ 2 , . . . )∈A inf i Cri {θi }, A ∈ Definition 9 ([11]): The random fuzzy variables ξ1 , ξ2 , . . . , ξn are
P(Θ). Then the set function Cr is a fuzzy measure on P(Θ) and independent if:
(Θ, P(Θ), Cr) is a credibility space (called the infinite product credi- 1) ξ1 (θ), ξ2 (θ), . . . , ξn (θ) are independent random variables for
bility space of (Θi , P(Θi ), Cri ), i = 1, 2, . . .). each θ;
IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 16, NO. 5, OCTOBER 2008 1381

2) E[ξ1 (·)], E[ξ2 (·)], . . . , E[ξn (·)] are independent fuzzy vari- space defined by
ables. ∞
Definition 10 ([11]): The random fuzzy variables ξ and η are iden- Θ= (Θi , Θi ) (16)
tically distributed if i= 1

and
sup inf {Pr{ξ(θ) ∈ B}} = sup inf {Pr{η(θ) ∈ B}} (12)
C r{A }≥α θ ∈A C r{A }≥α θ ∈A
Cr{A} = sup inf min(Cri {θi }, Cri {θi })
((θ 1 , θ 1 ), (θ 2 , θ 2 ), . . . )∈A i
for any α ∈ (0, 1] and Borel set B of real numbers.
Definition 11 ([31]): The random fuzzy variables ξi , i ∈ I are said to
for any A ∈ P(Θ).
be i.i.d. if and only if ξi 1 , ξi 2 , . . . , ξi n are i.i.d. random fuzzy variables
For any fixed θ, E[ξi (θ)], E[ηi (θ)], and E[C(t)(θ)] represent the
for all finite collections {i1 , i2 , . . . , in } of I, where I is an index set.
expected values of the random variables ξi (θ), ηi (θ), E[N (t)(θ)],
and C(t)(θ), respectively. When θ is varied over all of
III. RANDOM FUZZY RENEWAL REWARD PROCESSES Θ, E[ξi (θ)], E[ηi (θ)], E[N (t)(θ)], and E[C(t)(θ)], as functions of
Let ξi denote the interarrival times between the (i − 1)th and θ ∈ Θ, are fuzzy variables. The α-pessimistic and α-optimistic values
ith events, i = 1, 2, . . . , respectively. Suppose ξi ’s are random of the fuzzy variables E[ξi (θ)], E[ηi (θ)], E[N (t)(θ)], and E[C(t)(θ)]
fuzzy variables, each defined on their respective credibility spaces will then play an important role in random fuzzy renewal processes.
(Θi , P(Θi ), Cri ). Let S0 = 0 and Such values are obtained by

 
E[ξi (θ)]α = inf x|µE [ξ i (θ )] (x) ≥ α (17)
Sn = ξ1 + ξ2 + · · · + ξn ∀ n ≥ 1. (13)
 
E[ξi (θ)]α = sup x|µE [ξ i (θ )] (x) ≥ α (18)
Also, let N (t) denote the total number of events that have occurred by  
E[ηi (θ)]α = inf x|µE [η i (θ )] (x) ≥ α (19)
time t. Then
 
   E[ηi (θ)]α = sup x|µE [η i (θ )] (x) ≥ α
N (t) = max n  0 < Sn ≤ t
(20)
(14)
n ≥0  
E[N (t)(θ)]α = inf x|µE [N (t )(θ )] (x) ≥ α (21)
which is hereafter referred to as a random fuzzy renewal variable.  
E[N (t)(θ)]α = sup x|µE [N (t )(θ )] (x) ≥ α (22)
Between the (i − 1)th and ith events, i = 1, 2, . . . , a random fuzzy
reward ηi , defined on the credibility space (Θi , P(Θi ), Cri ), is inter-
 
E[C(t)(θ)]α = inf x|µE [C (t )(θ )] (x) ≥ α (23)
preted as the reward earned at the end of the ith renewal cycle. Here, the  
credibility spaces (Θi , P(Θi ), Cri ) and (Θi , P(Θi ), Cri ) are allowed E[C(t)(θ)]α = sup x|µE [C (t )(θ )] (x) ≥ α . (24)
to be different.
In general, suppose: Theorem 1: Let (ξ1 , η1 ), (ξ2 , η2 ), . . . be a sequence of pairs of i.i.d.
a) the random fuzzy variables ξ1 , ξ2 , . . . are i.i.d. positive random positive random fuzzy interarrival times defined on the infinite prod-
fuzzy variables; uct credibility space (Θ, P(Θ), Cr), N (t) the random fuzzy renewal
b) the random fuzzy variables η1 , η2 , . . . are i.i.d. positive random variable as given by (14), and C(t) the total reward as defined by (15).
fuzzy variables; Then, for any α ∈ (0, 1]
c) the sequences {ξi } and {ηi } are mutually independent;
d) the image sets Fi of ξi are totally ordered with respect to stochas- E [C(t)(θ)]α = E [N (t)(θ)]α · E[η1 (θ)]α (25)
tic ordering, i.e., for any θ1 , θ2 ∈ Θi and r ∈
, either
E [C(t)(θ)]α =E [N (t)(θ)]α · E[η1 (θ)]α (26)
Pr{ξi (θ1 ) ≤ r} ≤ Pr{ξi (θ2 ) ≤ r}
where E [C(t)(θ)]α , E [C(t)(θ)]α , E [N (t)(θ)]α , E [N (t)(θ)]α ,
(denoted by ξi (θ2 ) ≤d ξi (θ1 )) E[η1 (θ)]α , and E[η1 (θ)]α are the α-pessimistic and α-optimistic
values of E[C(t)(θ)], E [N (t)(θ)], and E[η1 (θ)], respectively.
or Proof: For each θ ∈ Θ,

Pr{ξi (θ1 ) ≤ r} ≥ Pr{ξi (θ2 ) ≤ r} 

N (t )(θ )

E [C(t)(θ)] = E ηi (θ) . (27)

(denoted by ξi (θ1 ) ≤d ξi (θ2 )) i= 1

where i = 1, 2, . . .. From the independence of the random fuzzy sequences {ξi } and {ηi }, it
Let C(t) denote the total reward earned by time t. Then follows that the random variable N (t)(θ) is independent of the random
sequence {ηi (θ)}. Thus

N (t )
N (t )(θ )
C(t) = ηi (15) 
i= 1 Pr ηi (θ) ≥ r
i= 1
where N (t) is the random fuzzy renewal variable defined by (14). 
∞
Thus, C(t) is a random fuzzy variable defined on credibility space = Pr {N (t)(θ) = k} Pr {η1 (θ) + · · · + ηk (θ) ≥ r} . (28)
(Θ, P(Θ), Cr), where (Θ, P(Θ), Cr) is an infinite product credibility k=1
1382 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 16, NO. 5, OCTOBER 2008

Hence, Proof: The following results have been proved in [31]:


N (t )(θ ) E [N (t)(θ)]α 0 1
lim = (33)
E ηi (θ) t →∞ t E[ξ1 (θ)]α 0
i= 1

N (t )(θ ) and
+∞  E [N (t)(θ)]α 0 1
= Pr ηi (θ) ≥ r dr lim = . (34)
t →∞ t E[ξ1 (θ)]α 0
0 i= 1
Applying Theorem 1 to these results leads to

+∞ ∞

= Pr {N (t)(θ) = k} Pr {η1 (θ) + · · · + ηk (θ) ≥ r} dr E [C(t)(θ)]α E [N (t)(θ)]α · E[η1 (θ)]α

0
lim = lim
k=1 t →∞ t t →∞ t
∞
+∞ E[η1 (θ)]α
= Pr {N (t)(θ) = k} Pr {η1 (θ) + · · · + ηk (θ) ≥ r} dr = (35)
E[ξ1 (θ)]α
k=1 0
and

∞

= Pr {N (t)(θ) = k} (E[η1 (θ)] + · · · + E[ηk (θ)]) . (29) E [C(t)(θ)]α E [N (t)(θ)]α · E[η1 (θ)]α
lim = lim
k=1 t →∞ t t →∞ t
Applying Proposition 1 to the above leads to E[η1 (θ)]α
= . (36)
E[ξ1 (θ)]α


N (t )(θ )
The theorem is therefore proven.
E ηi (θ)
Now, let ξ be one of the fuzzy variables with the α-pessimistic
i= 1 α value E[ξ1 (θ)]α and the α-optimistic value E[ξ1 (θ)]α , and η be one

∞ of the fuzzy variables with the α-pessimistic value E[η1 (θ)]α and
= Pr {N (t)(θ) = k} (E[η1 (θ)] + · · · + E[ηk (θ)]) the α-optimistic value E[η1 (θ)]α , then the following theorem can be
k=1 α established.
Theorem 3 (Random fuzzy renewal reward theorem): Let (ξ1 , η1 ),

∞

= (Pr {N (t)(θ) = k} (E[η1 (θ)] + · · · + E[ηk (θ)]))α (ξ2 , η2 ), . . . be a sequence of pairs of i.i.d. positive random fuzzy
k=1
interarrival times defined on the infinite product credibility space
(Θ, P(Θ), Cr), N (t) the random fuzzy renewal variable as given by

∞
(14), and C(t) the total reward as defined by (15). If E[η/ξ] is finite,
= (Pr {N (t)(θ) = k})α (E[η1 (θ)] + · · · + E[ηk (θ)])α
then  
k=1 E[C(t)] η
lim =E . (37)

∞
t →∞ t ξ
= (Pr {N (t)(θ) = k})α (E[η1 (θ)]α + · · · + E[ηk (θ)]α )
Proof: It follows from Definition 8 and Proposition 2 that
k=1

 
E [C(t)(θ)]α E [C(t)(θ)]α
1

∞ E[C(t)]
=
1
+ dα (38)
= (Pr {N (t)(θ) = k})α · kE[η1 (θ)]α t 2 0 t t
k=1
 and  
1  

∞ η 1 E[η1 (θ)]α E[η1 (θ)]α
= k Pr {N (t)(θ) = k} · E[η1 (θ)]α E = 
+ dα. (39)
ξ 2 0 E[ξ1 (θ)]α E[ξ1 (θ)]α
k=1 α
Hence, it suffices to prove that
= E[N (t)(θ)]α · E[η1 (θ)]α . (30)


E [C(t)(θ)]α
1 1
E[η1 (θ)]α
lim dα = dα (40)
Hence, the result (25) holds. Assertion (26) can be established though t →∞
0 t 0 E[ξ1 (θ)]α
a similar proof procedure. Thus, the theorem is proven.
Theorem 2: Let (ξ1 , η1 ), (ξ2 , η2 ), . . . be a sequence of pairs of i.i.d.
and


E [C(t)(θ)]α
1 1
E[η1 (θ)]α
nonnegative random fuzzy interarrival times defined on the infinite lim dα = 
dα. (41)
t →∞ t 0 E[ξ1 (θ)]α
product credibility space (Θ, P(Θ), Cr), N (t) the random fuzzy re- 0

newal variable as given by (14), and C(t) the total reward as defined The following shows the proof for assertion (40), while assertion (41)
by (15). Then can be proven similarly.
First of all, it follows from Theorem 2 that
E [C(t)(θ)]α E[η1 (θ)]α
lim = (31) E [C(t)(θ)]α E[η1 (θ)]α
t →∞ t E[ξ1 (θ)]α lim = (42)
t →∞ t E[ξ1 (θ)]α
E [C(t)(θ)]α E[η1 (θ)]α
lim = (32) provided that E[ξ1 (θ)]α is continuous at point α, α ∈ (0, 1].
t →∞ t E[ξ1 (θ)]α What is now needed is to prove that there exists one integrable
provided that the α-pessimistic value E[ξ1 (θ)]α and the α-optimistic function g1 (α, t) such that
value E[ξ1 (θ)]α of the fuzzy variables E[ξ1 (θ)] are continuous at point E [C(t)(θ)]α
α, α ∈ (0, 1]. ≤ g1 (α, t). (43)
t
IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 16, NO. 5, OCTOBER 2008 1383

In order to do this, define a sequence of i.i.d. random variables It follows from (44) that E [ξ1 (θ  )] = E [ξ1 (θ)]α . Thereby,
{ξi (θ  )}, θ  ∈ Ai , such that
E[η1 (θ)]α E[η1 (θ)]α
ξi (θ) ≤d ξi (θ  )  ↓ , as M → ∞. (56)
(44) 
E ξ 1 (θ ) E [ξ1 (θ)]α
for any θ ∈ Ai , where
n Ai is the α-level set of E [ξi (θ)] , i = 1, 2, . . . . By the assumption that E[η/ξ] is finite, it follows that
Let Sn (θ  ) = i = 1 ξi (θ  ), and thus, {Sn (θ  ), n ≥ 1} is a stochas-
E[η1 (θ)]α /E [ξ1 (θ)]α is integrable. Thus, from the
 monotonic con-
tic renewal process with the renewal variable
vergence theorem (see [3]), E[η1 (θ)]α /E ξ 1 (θ  ) is integrable and

N (t)(θ  ) = max {n  Sn (θ  ) ≤ t}. (45)
1
1
n ≥0
2E[η1 (θ)]α 2E[η1 (θ)]α
  dα ↓ dα (57)
In this case, it is obvious that 0 E ξ 1 (θ  ) 0 E [ξ1 (θ)]α

E[N (t)(θ  )] = E[N (t)(θ)]α . (46) as M → ∞. Therefore, there exists g1 (α, t) = 2E[η1 (θ)]α /
 
  E ξ 1 (θ) α such that
Define a new stochastic renewal process S n (θ  ), n ≥ 1 by letting
E [C(t)(θ)]α

n
≤ g1 (α, t). (58)
S n (θ  ) = ξ i (θ  ) (47) t

Finally, since E [N (t)(θ)]α is an almost surely continuous function

i= 1

and    of α, α ∈ (0, 1], the dominated convergence theorem (again, see [3])
N (t)(θ  ) = max n  S n (θ  ) ≤ t (48) implies that (40) holds.
n ≥0
As indicated previously, assertion (41) can be similarly proven, and
where  hence, the theorem is proven.
ξi (θ  ), if ξi (θ  ) ≤ M Remark 3: If the random fuzzy variables ξi and ηi degenerate to
ξ i (θ  ) = (49)
M, if ξi (θ  ) > M random variables, then the result in Theorem 3 degenerates to the form
and M < t. Since
E[C(t)] E[η1 ]
lim = (59)
S N (t )(θ  )+ 1 (θ  ) ≤ t + M (50) t →∞ t E[ξ1 ]

it follows from Wald’s equation in stochastic sense (see [22]) that which is just the conventional result in the stochastic case [22]. Thus,
      the previous theorem is an extension of the corresponding stochastic
E N (t)(θ  ) + 1 · E ξ 1 (θ  ) ≤ t + M. (51) theorem.
Remark 4: If the random fuzzy variables ξi and ηi degenerate to
Thus fuzzy variables, then the result in Theorem 3 degenerates to the form
   
E N (t)(θ  ) + 1 t+M 2
≤  <  . (52) lim
E[C(t)]
=E
η1
(60)
t 
tE ξ 1 (θ ) E ξ 1 (θ  ) t →∞ t ξ1

Since S n (θ  ) ≤ Sn (θ  ), which is just the conventional result in the fuzzy case [29]. Thus, the
  previous theorem is an extension of the corresponding fuzzy theorem.
N (t)(θ  ) ≥ N (t)(θ  ) E N (t)(θ  ) ≥ E [N (t)(θ  )] .
and It is interesting to note that as with conventional stochastic renewal
(53) reward processes [22], if a renew cycle is completed every time a
By (46), (52), and (53), the following holds: renewal occurs, then Theorem 3 states that the long-run expected return
is just the expected return earned during a cycle, divided by the expected
E [C(t)(θ)]α E [N (t)(θ)]α · E[η1 (θ)]α time of that cycle.
=
t t
IV. APPLICATION EXAMPLE
E [N (t)(θ  )] · E[η1 (θ)]α
=
t To illustrate the utility of random fuzzy renewal award process mod-
(E [N (t)(θ  )] + 1) E[η1 (θ)]α eling, this section presents a simple, worked example. The main purpose
< of this exercise is to show how the essential mathematical concepts in-
t
troduced in the last section may be applicable to addressing a realistic
2E[η1 (θ)]α problem. Detailed computations regarding this example are omitted,
< .
E[ξ 1 (θ  )] while comprehensive real-world applications of the proposed model-
ing approach are beyond the scope of this short paper and remain as
Moreover, it follows from (49) that
active research.
  Consider an insurance company dealing with its claims. The inter-
E ξ 1 (θ  ) ↑ E [ξ1 (θ  )] , as M → ∞. (54) arrival times between the (n − 1)th and nth claims ξn can be modeled
by i.i.d. positive random fuzzy variables with the following exponential
That is, distribution 
E[η1 (θ)]α E[η1 (θ)]α 1 − e−x / u n , if 0 ≤ x < ∞
 ↓ , as M → ∞. (55)
E ξ 1 (θ  ) E [ξ1 (θ  )] 0, otherwise,
1384 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 16, NO. 5, OCTOBER 2008

where un are i.d.d. fuzzy variables n = 1, 2, . . . . Also, the claim sizes renewal processes [24], age-dependent branching processes, delayed
ηn can be represented as i.i.d. positive random fuzzy variables with the renewal processes, and stationary point processes. Jointly, such exten-
following exponential distribution: sions would enable the modeling of a variety of real-world application
 problems.
1 − e−x / v n , if 0 ≤ x < ∞
0, otherwise
ACKNOWLEDGMENT
where vn are i.i.d. fuzzy variables. Moreover, {ξn } and {ηn } may be
The authors are grateful to the members of the project team for help-
assumed to be mutually independent.
ful discussions, but will take full responsibility for the views expressed
Let N (t) denote the total number  of claims that
 have occurred in this short paper. They also thank the anonymous reviewers for their
by time t, where N (t) = maxn ≥0 n  0 < Sn ≤ t . Then, C(t) =
N (t ) positive and constructive comments.
i= 1
ηi represents the total amount of the claims accumulated up
to time t, and the process {C(t), t ≥ 0} is a random fuzzy renewal
reward process. Thus, it follows from Theorem 3 that the average REFERENCES
expected claim size in the long run can be calculated as [1] G. Allan, K. Oleq, and S. Josef, “Equivalences in strong limit theorems
E[C(t)]
v  for renewal counting processes,” Stat. Prob. Lett., vol. 35, pp. 381–394,
1 1997.
lim =E .
t →∞ t u1 [2] G. Alsmeyer, “Superposed continuous renewal processes: A Markov re-
newal approach,” Stochastic Process. Their Appl., vol. 61, pp. 311–322,
As a particular instance of this application problem, for computa- 1996.
tional simplicity, assume that all values of the fuzzy variables are rep- [3] J. E. Derwent, Measure Theory, From MathWorld—A Wolfram Web Re-
resented in triangular form. Let the number of days u1 be (20, 25, 30), source, created by E. W. Weisstein [Online]. Available: http://mathworld.
and the claim amount v1 be (80, 100, 120) in British pounds sterling. wolfram.com/MeasureTheory.html
[4] D. Dubois and H. Prade, Fuzzy Sets and Systems: Theory and Applications.
By Definition 2 New York: Academic, 1980.
[5] J. A. Ferreira, “Pairs of renewal processes whose superposition is a renewal
v1 α = 80 + 20α v1α = 120 − 20α process,” Stochastic Process. Their Appl., vol. 86, pp. 217–230, 2000.
[6] C. M. Hwang, “A theorem of renewal process for fuzzy random variables
u1 α = 20 + 5α u1 α = 30 − 5α. and its application,” Fuzzy Sets Syst., vol. 116, pp. 237–244, 2000.
[7] H. Kwakernaak, “Fuzzy random variables—I,” Inf. Sci., vol. 15, pp. 1–29,
Then 1978.
v 
1   [8] H. Kwakernaak, “Fuzzy random variables—II,” Inf. Sci., vol. 17, pp. 253–
1 1 v1 α v  278, 1979.
E = + 1 α dα
u1 2 0 u1 α u1 α [9] S. Li, R. Zhao, and W. Tang, “Fuzzy random homogeneous poisson process
and compound poisson process,” J. Inf. Comput. Sci., vol. 1, pp. 207–224,
= 20(ln 3 − ln 2) − 4. 2006.
[10] S. Li, R. Zhao, and W. Tang, “Fuzzy random delayed renewal process and
The average expected claim size in the long run can therefore be com- fuzzy random equilibrium renewal process,” J. Intell. Fuzzy Syst., vol. 18,
pp. 207–224, 2006.
puted as
[11] X. Li and B. Liu, “New independence definition of fuzzy random variable
and random fuzzy variable,” World J. Model. Simul., vol. 2, pp. 338–342,
E[C(t)]
lim = 20(ln 3 − ln 2) − 4 ≈ 4.11. 2006.
t →∞ t [12] B. Liu, Theory and Practice of Uncertain Programming. Heidelberg,
Germany: Physica-Verlag, 2002.
In other words, the proposed modeling approach leads to the result that [13] B. Liu, “A survey of credibility theory,” Fuzzy Optim. Decis. Making,
under the above-assumed distributions, the insurance company may vol. 5, pp. 387–408, 2006.
expect to receive claims at the rate of £4.11 per day. [14] B. Liu, “Toward fuzzy optimization without mathematical ambiguity,”
Fuzzy Optim. Decis. Making, vol. 1, pp. 43–63, 2002.
[15] B. Liu, Uncertainty Theory: An Introduction to Its Axiomatic Foundations.
V. CONCLUSION Berlin, Germany: Springer-Verlag, 2004.
This short paper has proposed a method to describe the renewal re- [16] B. Liu and Y. Liu, “Expected value of fuzzy variable and fuzzy expected
value models,” IEEE Trans. Fuzzy Syst., vol. 10, no. 4, pp. 445–450, Aug.
ward processes involving random fuzzy interarrival times and rewards. 2002.
The approach takes advantage of i.i.d. nonnegative random fuzzy vari- [17] B. Liu, “A survey of entropy of fuzzy variables,” J. Uncertain Syst., vol. 1,
ables via considering the α-pessimistic and α-optimistic values of such pp. 4–13, 2007.
variables. In doing so, important properties of random fuzzy renewal [18] Y. Liu and B. Liu, “Expected value operator of random fuzzy variable
and random fuzzy expected value models,” Int. J. Uncertainty, Fuzziness
reward processes, including the theorem on random fuzzy elemen- Knowl.-Based Syst., vol. 11, pp. 195–215, 2003.
tary renewal rewards, are established. This short paper generalizes the [19] Y. Liu and J. Gao, “The independence of fuzzy variables in credibility
mechanism for modeling conventional stochastic renewal processes. theory and its applications,” Int. J. Uncertainty, Fuzziness Knowl.-Based
The results of this research are currently being applied to address Syst., vol. 15, pp. 1–20, 2007.
the problem of risk and reliability analysis, in the area of terrorism risk [20] S. Nahmias, “Fuzzy variables,” Fuzzy Sets Syst., vol. l, pp. 97–110, 1978.
[21] E. Popova and H. C. Wu, “Renewal reward processes with fuzzy rewards
management. This is rooted in the observation that the loss brought and their applications to T -age replacement policies,” Eur. J. Oper. Res.,
by a terrorist attack may be represented by a random fuzzy variable. vol. 117, pp. 606–617, 1999.
Therefore, the average loss per time unit may be evaluated by ap- [22] S. M. Ross, Stochastic Processes. New York: Wiley, 1996.
plying the theorem on random fuzzy renewal rewards, as developed [23] G. Shafer, A Mathematical Theory of Evidence. Princeton, NJ: Princeton
Univ. Press, 1976.
in the short paper. This ongoing investigation will help better reveal [24] Q. Shen, R. Zhao, and W. Tang, “Random fuzzy alternating renewal pro-
the potential, and also the limitations, of the proposed approach. Fi- cesses,” Soft Comput., to be published.
nally, it is worth noting that the ideas presented here can be extended [25] D. Stoyan, Comparison Methods for Queues and Other Stochastic Models.
similarly to dealing with other renewal processes such as alternating New York: Wiley–Interscience, 1983.
IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 16, NO. 5, OCTOBER 2008 1385

[26] K. Takis and L. Gunter, “On the use of Lyapunov function methods in [29] R. Zhao and B. Liu, “Renewal process with fuzzy interarrival times and
renewal theory,” Stochastic Process. Their Appl., vol. 79, pp. 165–178, costs,” Int. J. Uncertainty, Fuzziness Knowl.-Based Syst., vol. 11, pp. 573–
1999. 586, 2003.
[27] G. Visaggio, “Assessment of a renewal process experimented in the field,” [30] R. Zhao and W. Tang, “Some properties of fuzzy random processes,”
J. Syst. Softw., vol. 45, pp. 3–17, 1999. IEEE Trans. Fuzzy Syst., vol. 14, no. 2, pp. 173–179, Apr. 2006.
[28] L. Zadeh, “Fuzzy sets as a basis for a theory of possibility,” Fuzzy Sets [31] R. Zhao, W. Tang, and H. Yun, “Random fuzzy renewal process,” Eur. J.
Syst., vol. 1, pp. 3–18, 1978. Oper. Res., vol. 169, pp. 189–201, 2006.