Академический Документы
Профессиональный Документы
Культура Документы
Information Sciences
journal homepage: www.elsevier.com/locate/ins
Data Mining and Optimization Research Group, Centre for Articial Intelligence Technology, Universiti Kebangsaan Malaysia, Bangi, Selangor, Malaysia
Department of Computer and Software Engineering, University of Birjand, Birjand, Iran
a r t i c l e
i n f o
Article history:
Received 20 March 2013
Received in revised form 31 December 2013
Accepted 5 March 2014
Available online 25 March 2014
Keywords:
Fuzzy job-shop scheduling
Review
Meta-heuristic algorithm
a b s t r a c t
Fuzzy job-shop scheduling problems (Fuzzy JSSPs) are a class of combinational optimization
problems known as non-deterministic polynomial-hard problems. In recent decades, a
number of researchers have expanded the theoretical models of Fuzzy JSSPs and introduced
algorithms to solve them. This paper reviews the classication of Fuzzy JSSPs, constraints
and objectives investigated in Fuzzy JSSPs, and the methodologies applied in solving Fuzzy
JSSPs. The paper centers on reviewing meta-heuristic algorithms as state-of-the-art
algorithms proposed for Fuzzy JSSPs. These algorithms are analyzed in three steps, namely,
pre-processing, initialization procedures, and improvement algorithms. Finally, possible
suggestions for future studies are obtained from this survey.
2014 Elsevier Inc. All rights reserved.
1. Introduction
Promptly addressing a job-shop scheduling problem (JSSP) is an important decision in the industry, economy, and
management. This problem is a class of combinational optimization problems known as non-deterministic polynomial-hard
problems [34]. The assumption that the duration times of JSSP in real-world problems have crisp value is often violated in
practice. Human-centered factors are incorporated into JSSPs; thus, considering fuzzy processing time due to man-made factors and fuzzy due date, which tolerates a certain amount of delay in the due date, may be more appropriate. Therefore, a
new type of JSSP, Fuzzy JSSP [9,33], has gained prominence. Numerous researchers have expanded the theoretical models
of crisp JSSPs, introduced new algorithms to solve them, and published thousands of papers in this domain from the mid1960s. However, the domain of Fuzzy JSSP was rst offered in 1995. Less than 60 papers are currently available in this domain. The diversity of investigated algorithms for solving Fuzzy JSSPs has also been unsatisfactory, indicating that research
on Fuzzy JSSPs remains in the infancy stage. This aspect motivates the present paper to gather and review the available literature on Fuzzy JSSPs for the rst time. This paper has three objectives: (1) to collect available information and background
on the Fuzzy JSSP domain; (2) to identify probable gaps in this domain; and (3) to encourage researchers to ll the identied
gaps in future studies.
This paper is organized as follows. Section 2 presents the overview and classication of Fuzzy JSSPs. Section 3 discusses
the variation in constraints and objectives investigated in Fuzzy JSSPs. Section 4 describes the exact and heuristic algorithms
applied for Fuzzy JSSPs. Section 5 reviews the three steps employed in meta-heuristic algorithms, namely, pre-processing,
initialization procedures, and improvement algorithms. Section 6 explains the benchmark datasets of Fuzzy JSSPs. Finally,
Section 7 summarizes this review and highlights the effective features for future studies.
381
382
Table 1
Parameters of the eight models of Fuzzy JSSPs.
Model
Constants
Criterion to approximate
fuzzy numbers
Objective function
Applicable constraints
(see Section 3)
Fuzzy double
number
Constraints (1)(11)
Sakawas criterion
Sakawas criterion
Leis criterion
Leis criterion
Fuzzy double
number
Fuzzy trapezoid
number
Fuzzy double
number
Sakawas criterion
Processing time
Due date
Xies model
Crisp
Lins model
Fuzzy triangular
number
Fuzzy triangular
number
Ghrayebs
model
Rodriguezs
model
Lei1s model Fuzzy JSSP
Fuzzy
Flexible JSSP
Sakawas
model
Songs
model
Lei2s model
Fuzzy triangular
number
Fuzzy triangular
number
Fuzzy triangular
number
Fuzzy triangular
number
Fuzzy triangular
number
Fuzzy triangular
number
Sakawas criterion
Leis criterion
Constraints (1)(11)
Constraints (1)(11)
Constraints (1)(11)
Constraints (1)(11)
Constraints, (2)(6),
(8)(11)(13)
Constraints (1)(11)
Constraints (1)(11)
Constraints (1)(11)
fuzzy numbers, the objective functions, and the constraints of these models are presented for ease of comparison. The definitions of the constraints are explained in the next section.
Xies model [89] is classied as Fuzzy JSSP with fuzzy due date. The processing, starting, completion times, the constraints
and solution space are similar with the crisp JSSP. Since the due date of a given job is not strict, a fuzzy due date is associated
e i d1 ; d2 ). The membership function of a
to this job. This fuzzy due date is as an interval on the positive part of real line ( D
i
i
fuzzy due date assigned to a given job (see Fig. 4) denotes the satisfaction degree of a decision maker for the completion time
1
2
2
1
of the job. The membership function of Ji is specied by di and di where i = 1, 2, . . . , n, di represents the latest due date and di
denotes the earliest due date for Ji. The objective function of Xies model is considered as maximizing the minimum degree of
satisfaction (fi(Cim)) over given job (Ji) and symbolized by SDmin. In mathematics, the formula of the degree of satisfaction (the
membership function of Ji is shown in Fig. 4) is presented as follow:
8
>
1
>
>
<
fi C im
>
>
>
:
383
C im 6 di
d2i C im
d2i d1i
di 6 C im 6 di
C im P
2
di
where Cim is completion time of the last operation of Ji. Also the formula of the objective function of this Fuzzy JSSP is denoted as follow:
MaximizeSDmin Maximizemin fi C im
16i6n
In fuzzy models of Fuzzy JSSPs with fuzzy processing time, although the constraints are similar with the crisp JSSP, uncertainty in processing times is associated with the operations. The triangular fuzzy numbers (TFN) are considered as fuzzy proe ij or P
e ij ) of a given operation (Oij) includes three
cessing times, which are shown Fig. 5. The fuzzy processing time ( PO
parameters: the optimistic value (P 1ij ), the most plausible value (P2ij ) and the pessimistic value (P 3ij ).
The membership of fuzzy processing time is also linear and presented as follow:
lp~ij t
8
>
0
>
>
>
>
< tP1ij
>
>
>
>
>
:
P 2ij P 1ij
P3ij t
P 3ij P 2ij
P 2ij 6 t 6 P3ij
Lins model [14,48,49,66,87] consists of fuzzy processing time and no due date. In this model, fuzzy processing time is
de-fuzzed using the mean value of a fuzzy number (Eq. (4)); the de-fuzzed makespan is subsequently minimized.
Pij P 2ij
1 3
Pij P2ij P2ij P1ij
4
Ghrayebs model [19,20,58,83] is classied as Fuzzy JSSP with fuzzy processing time. This model aims to identify an
optimal fuzzy schedule by minimizing both the total integral value (in Eq. (5), b e [0, 1]) and the uncertainty of the fuzzy
makespan which are formulated as in Eqs. (5) and (6), respectively. Sakawas criterion is applied to approximate the fuzzy
max operation.
e max
TIV b C
1
bC 1max C 2max 1 bC 3max
2
e max C 3 C 1
U C
max
max
5
6
Rodriguezs model [4547,64,71,72,74,91] is classied as Fuzzy JSSP with fuzzy processing time. Similar to Ghrayebs
model, Sakawas criterion is used to approximate the fuzzy max. However, its objective is to determine a fuzzy schedule that
will minimize the expected fuzzy makespan and is formulated as follows:
e max
E C
384
Lei1s model [37,40,41,43,8486,97] is classied as Fuzzy JSSP with fuzzy processing time. Leis criterion is used to
approximate the fuzzy max to generate the fuzzy schedules. The fuzzy problems in this model are categorized into two main
groups, namely, Fuzzy JSSP and Fuzzy Flexible JSSP. Each operation processes on only one machine in Fuzzy JSSPs [40,84],
e max ). Fuzzy Flexible JSSPs [37,41,43,85,86,97] allow an operaand aims to minimize the expected fuzzy makespan (min E C
tion to be processed by one machine out of a set of candidate machines. The problem of fuzzy exible job-shop scheduling is
to assign each operation to an appropriate machine and to sequence the operations on the machines in order to minimize
e max ) as its objective function.
fuzzy makespan (min C
Sakawas model [11,17,28,42,44,51,63,65,6870,73,7577,98] is classied as Fuzzy JSSP with both fuzzy processing time
and fuzzy due date. Fuzzy double and triangular numbers represent the fuzzy due date and fuzzy processing time, respectively. Sakawas criterion is used to approximate the fuzzy max. The objective function of this model is to maximize the mine im ) of the last
imum of the agreement index (max AImin). The AI presents the portion of the fuzzy completion time ( C
e i ), as the standard criterion in this class of Fuzzy JSSP.
operation of Ji that has been completed by the fuzzy due date ( D
AIi
e im \ D
e i
area C
e
area C im
leD t
i
8 1
tdi
>
>
>
d2i d1i
>
>
>
>
<1
d4i t
>
>
>
>
d4i d3i
>
>
>
:
0
di 6 t 6 di
di 6 t 6 di
di 6 t 6 di
1
Lei2s model [27,35,36,38,39] is classied as Fuzzy JSSP with both fuzzy processing time and fuzzy due date. In this model,
the fuzzy processing time and fuzzy due date as well as the objective function are similar to those in Sakawas model. Meanwhile, the fuzzy max is approximated using Leis criterion.
Fig. 8. AIi during which a triangular fuzzy processing time meet a trapezoid fuzzy due date.
385
386
crisp criteria instead of the vague criterion. The total integral value of the fuzzy makespan is a convex combination of the
right and left integral values through an index of optimism. The left and right integrals are used to reect the optimistic
and pessimistic viewpoints of the fuzzy makespan, respectively. The uncertainty of the fuzzy makespan can also be measured by its spread.
Minimize the expected fuzzy makespan: This objective function was proposed for optimizing Fuzzy JSSPs with fuzzy proe matches the neutral scalar replacement of
cessing time. Given that the expected value of a triangular fuzzy number (E A)
a fuzzy interval and can be obtained as the center of gravity of its mean value, [4547,64,71,72,74,91] considered minimizing
e max ).
this criterion of fuzzy makespan (min E C
Maximize the minimum of the agreement index: This objective function optimizes the portion of fuzzy completion time that
is completed by fuzzy due date as the standard criterion for Fuzzy JSSPs with both fuzzy processing time and fuzzy due date
[17,27,36,42,44,51,65,68,69,75,80,90,98]. The agreement index can be regarded as the extended version of tardiness or earliness in the fuzzy case to describe the degree of customers satisfaction with the delivery.
4. Exact and heuristic methods applied to Fuzzy JSSPs
Although the history of crisp JSSP solution can be traced back to the mid-1960s when several exact methods were proposed, the history of Fuzzy JSSPs has only started in the mid-1990s when branch and bound algorithms [33] were applied for
a small size of Fuzzy JSSPs. The characterization of the methodologies applied to solve Fuzzy JSSPs is presented in Fig. 11.
Unfortunately, only one paper used an exact method to solve the Fuzzy JSSP [33] because exact methods are able to identify
an optimal solution for every small instance of Fuzzy JSSP in bounded time, and methods for solving real large-scale application of Fuzzy JSSPs in polynomial time are unavailable. Therefore, exact algorithms need exponential computation time,
specically in cases in which these problems cause an impractical computational burden. Integer programming [53], one
of the earliest works on exact methods, has not been applied for Fuzzy JSSPs due to the abovementioned causes; nevertheless, [1,12,13,24,54] applied this method for crisp JSSPs.
The implementation of exact methods for solving JSSPs often causes unreasonable computational times. This fact has
motivated most researchers on Fuzzy JSSPs to employ approximation methods. Although approximation methods cannot
387
be guaranteed to identify the optimal solutions, these methods can derive near-optimum solutions within reasonable computational times. Approximate algorithms are classied into heuristic [22] and meta-heuristic algorithms [23]. The term
heuristic is derived from the Greek verb heuriskein (et.irjeim), which means to nd, and the sufx meta means
beyond in an upper level [3]. It means heuristic algorithms refer to experience-based techniques and meta-heuristic algorithms refer to high-level iterative techniques which guide a subordinate heuristic by using intelligent concepts to explore
the solution space and exploit a solution close to the optimal solution.
The rst person who applied heuristic for Fuzzy JSSPs was Lin [48,49] who considered a Fuzzy JSSP with fuzzy processing
time. He de-fuzzed the problem and solved its corresponding crisp JSSP using Johnsons construction algorithm. The main
disadvantage of the heuristic is its limitation in solving JSSPs with more than two machines. Ramkumar et al. [67] developed
a fuzzy rule-based system that determines the study to be allocated to N number of machines with M number of jobs for
Fuzzy JSSPs with both fuzzy processing time and fuzzy due date. Although numerous heuristic methods have been proposed
based on dispatching rules for crisp JSSPs [2,5,6,8,21,26,52,55,60,78,82], only one implementation of fuzzy priority dispatching rules for Fuzzy JSSP exists because dispatching rules cannot perform effectively over complex Fuzzy JSSPs and have limited use in practical JSSPs. Notably, dispatching rules that have been applied on crisp JSSPs have the advantages of simplicity
of implementation and ability to identify a reasonable solution in less computational time. Wu et al. [87] solved Fuzzy JSSP
with both fuzzy processing time by de-fuzzying the problem and applying shifting bottleneck procedure to solve the defuzzed JSSP. This algorithm includes optimizing a one-machine scheduling sub-problem for each machine (work-center).
The effectiveness of this heuristic relies on the ability to efciently obtain good schedules for individual work-centers and
to accurately model the interactions between work-centers. The shifting bottleneck heuristic offers several advantages over
the exact methods and dispatching rules. First, it decomposes the shop into a number of disjoint work-centers, which allows
each work-center to be scheduled using the most suitable procedure. Second, it uses the global information instead of the
local information used by dispatching rules. Both applications of heuristics on Fuzzy JSSPs were implemented for small
de-fuzzed JSSPs and were not solved for the fuzzy benchmark datasets proposed by [76]. Therefore, these applications for
Fuzzy JSSPs cannot be compared with other methods. Nevertheless, the general properties of the heuristics for crisp JSSPs
are presented in the above paragraphs.
5. Meta-heuristic algorithms
The idea of meta-heuristics algorithms was proposed by [23]. Several denitions have been subsequently presented for
these algorithms. A meta-heuristic is formally dened as an iterative generation process which guides a subordinate heuristic by
combining intelligently different concepts for exploring and exploiting the search space, learning strategies are used to structure information in order to nd efciently near-optimal solutions [59]. Based on the denition published in [3, p. 270], meta-heuristics are
388
intelligent means of exploring the solution space and exploiting the optimal solution. Implementing one meta-heuristic for
Fuzzy JSSPs involves three steps, namely, pre-processing, initialization, and improvement (see Fig. 12). Two tasks are dened
for pre-processing, namely, representing or encoding the solution space and designing a decoder algorithm to generate feasible
fuzzy schedule corresponding to the encoded point. Subsequently, an initial population or initial single point needs to be generated in the initialization. Finally, the initial solutions are improved by intelligent algorithms based on exploration and exploitation techniques on the solution space of Fuzzy JSSPs. The characterization of meta-heuristic algorithms for Fuzzy JSSPs is
shown in Fig. 12.
5.1. Pre-processing
One of the key issues in the successful application of meta-heuristics to JSSPs is the means through which a schedule
should be encoded to a point in search space; that is, the suitable selection of an encoding scheme is extremely important
in enhancing the search effectiveness of any meta-heuristic. Therefore, nine representations for JSSPs have been proposed in
the past 25 years. The denitions and properties of the representation schemes for JSSPs are described in [7]. These representations are classied into two general encoding approaches, namely, direct and indirect approaches. In the direct approach, a schedule (a point of the solution space) is encoded into a permutation style (a point of the search space), and
meta-heuristic algorithms are applied to evolve these points to determine a better schedule. Representations such as
job-based, operation-based, job pair relation-based, completion time-based, and random key representations belong to this
category. In the indirect approach, such as priority rule-based representation, a sequence of dispatching rules for job assignment, instead of a schedule, is encoded into the search space, and meta-heuristic algorithms are used to evolve these points
to identify a better sequence of dispatching rules. A schedule is then constructed with the sequence of dispatching rules.
Preference list-based, priority rule-based, disjunctive graph-based and machine-based representations belong to this category. The percentage of these representations applied in the statistical population for Fuzzy JSSPs is illustrated in Fig. 13.
Note that these percentages are taken based on fty four reviewed papers as presented in Appendix A.
All of the encoding schemes proposed for JSSPs can generate an active schedule using the decoder. Cheng et al. [7] classied the degree of complexity of the decoder into four levels: (Level 0) No decoder; (Level 1) Simple mapping relation; (Level
2) Simple heuristic; and (Level 3) Complex heuristic. The advantages and disadvantages of the encoding schemes as well as the
complexity level of the decoder applied for each encoding scheme are presented in Table 2. In this table, n is number of jobs
and m is number of machines.
Operation-based representation [84] encodes a schedule into a sequence of operations. However, all possible permutations
of these sequences cannot dene feasible schedules, resulting in the existence of the precedence constraints. An alternative
was introduced, in which all operations for a job are denoted with the same character and then interpreted based on the
order of appearance in the sequence [62]. The application of this representation has the rst-rank position among the other
representations [1720,27,40,4447,58,70,71,79,80,84,90,91,98]. Moreover, the decoder of this representation is engaged in
two main steps. First, the encoded point is translated to a list of ordered operations. Second, the schedule is generated by a
one-pass heuristic based on the list. The Gifer and Thompson (G&T) algorithm by [21] is a well-known constructive procedure that has been used as a decoder for this representation on the crisp JSSP domain. An extended version of the G&T
algorithm to generate fuzzy schedule based on fuzzy operators was utilized as a fuzzy decoder by
389
Fig. 13. Slice of encoding schemes applied in the statistical population of the current review for Fuzzy JSSPs.
Table 2
Encoding schemes and the complexity of their decoders for JSSPs.
Direct representations
Indirect representations
Encoding
Example
researches
Advantages
Operationbased
[19,58,71,84]
Job-based
[30,32]
Job pair
relationbased
[57,88]
Completion
time-based
[48,73,75,76]
Random
keys
[36,38,39]
Preference
list-based
[33,63]
Priority
rule-based
[11,35,42]
Disjunctive
graphbased
Machinebased
Search
space
Decoding
(nm)!
Simple
mapping
relation
n!
Simple
mapping
relation
Simple
mapping
relation
(nm)!
Simple
mapping
relation
(n!)m
Simple
heuristic
(n!)m
Simple
heuristic
Identify a sequence of
dispatching rules for job
assignment
The simplicity of the structure
[14,63,64,72,74,87] Have visual representation of
relationships among operations
[11]
Disadvantages
(2nm)m
(nm)1 No
decoder
Complex
heuristic
Complex
heuristic
[17,27,44,70,71,79,80,90,91]. Another decoder for this representation was proposed by [16] and extended for the Fuzzy JSSP
domain by [1820].
Job-based representation [25] includes a list of n jobs (the sequence of jobs). This encoding scheme rst schedules all operations of the rst job on the list, then the operations of the second job on the list, and so forth until all of the operations of the
remaining jobs are scheduled. This scheme is not being used in the Fuzzy JSSP domain up until now.
Preference list-based representation [33,63] divides an encoded point into m sub-points, where each point belongs to one
machine. Each sub-point comprises a string of operations, which has to be processed on the related machine. Sub-points do
not explain the operating sequence on the machine. They are the preference lists, where each machine has its own preference
list.
Job pair relation-based representation [89] comprises a binary matrix, which determines the precedence relation of two
jobs in corresponding machines.
390
Priority rule-based representation [11,35,42] includes the dispatching of a sequence of rules for job assignment. Its decoder
constructs a schedule using a priority dispatching heuristic. The G&T algorithms [21] are considered as the basis of all priority rule-based heuristics.
Disjunctive graph-based representation [14,63,64,72,74,87] is a visual job pair relation-based representation. In the disjunctive graph, G = (N, A, E), N nodes represent all of the operations, A represents the arcs that connect the consecutive operations of the same jobs, and E represents the disjunctive arcs that connect the operations that are yet to be processed by the
same machine. A decoder for this representation should translate the orientations of all disjunctive arcs into the sequences of
operations on their corresponding machines before constructing their schedules.
Completion time-based representation was proposed by [48,49,68,69,73,7577,83]. This encoding scheme assorts the completion times of operations without a decoder. Although this representation is unsuitable for most meta-heuristics, its application on the Fuzzy JSSP domain is of second priority because the construction of a fuzzy schedule requires complex
calculation operators.
Machine-based representation consists of a sequence of machines, and the procedure of its decoder is based on a shifting
bottleneck heuristic on the sequence. Fayad and Petrovic [11] combined this representation with the priority rule-based representation to encode the Fuzzy JSSP. Lei [37] merged the machine- and operation-based representations to serve as the solution space of the Fuzzy JSSP.
Random key representation [36,38,39] encodes a Fuzzy JSSP solution with random numbers. Each encoded point consists of
two parts, namely, an integer part in set {1, 2, . . . , m}, and a fraction part that is randomly generated from (0, 1). The integer
part is interpreted as the machine assignment for that job and sorted the fractional parts that provide the job sequence on
each machine. A sequence of all operations, similar to the operation-based representation, is considered as the integer part of
their random key representation.
5.2. Initialization
Initialization algorithms are typically the fastest approximate methods, but they often return solutions that are inferior
compared to those of improvement algorithms. Initialization of JSSP has been carried out using various methods, such as
random methods, priority rules, and heuristic algorithms. The high quality of the initial population speeds up the meta-heuristic algorithms [15,61], although researchers are less concerned with the production of an initial population than the other
steps of the meta-heuristic algorithm. Random initializations are typically the most preferred procedure. Fig. 14 shows the
application percentages of the initialization procedures in the statistical population for the Fuzzy JSSPs. Again, these percentages are based on fty four reviewed papers as in Appendix A.
Previous literature has demonstrated that most researchers have used random techniques to generate the initial population. [44,68,69,73,7577] extended the G&T to generate feasible fuzzy schedules, which are then embedded with random
selections to produce initial fuzzy schedules. Xu et al. [91] randomly initialized the antibodies to generate the initial population of their algorithm. The random generation of population size initial individuals for genetic algorithm was considered
by [1720,36,38,39,41,47,71,79,84,90]. Niu et al. [58] and Lei [40] initialized a population of particles with random positions
and velocities. Lei [43] randomly generated an initial swarm (S) using the N solution. Song et al. [80] and Zheng and Li [98]
generated initial solutions using random food sources. Given that the initial population is randomly produced in most metaheuristic algorithms, it does not only require a longer time to compute for the optimal solution of Fuzzy JSSP, but also decreases the possibility of nding an optimal solution. It may also cause certain encoding schemes to generate illegal encoded
points. The main advantages of random initializations, which have attracted the interest of researchers, are their simple
implementation, their very short computational time, and the ne diversication of the initially generated points in the
search space of the problem.
[11,35,42,63] used priority rules as their second strategy to generate initial fuzzy schedules. Priority rules are easy to execute, have few requirements for computational power, and have less time complexity. Each priority rule in the initialization
can only generate one solution. To overcome this limitation, several researchers [45,46] combined a set of priority rules with
random selections to produce the initial population in their algorithms.
Unfortunately, no heuristic initialization procedures on the Fuzzy JSSP domain exist, because such procedures have become less interesting to these heuristics, which often have complex structures and a computational time that is longer than
the random techniques and priority rules processes. With regard to the application of the heuristic initialization procedure
on the crisp JSSP domain [4,10,31,92,9496], the signicant advantage of the heuristic initialization procedures is their ability to generate an initial population that is close to the optimal solutions. This advantage enables the improvement algorithms to quickly reach an optimal solution and compensate for the time spent on the initialization. These heuristics are
the fastest approximation algorithms to solve JSSP. Their solutions, instead of the optimal solutions with minute deviation,
can also be considered.
5.3. Improvement
The last part of the meta-heuristic for Fuzzy JSSP is the improvement algorithms, which are initiated from some initial
solutions to nd better solutions by exploring and exploiting the search space of the problem. Several intelligent approaches,
such as genetic algorithm (GA), particle swarm optimization (PSO), ant colony optimization (ACO), neural networks (NN),
391
Fig. 14. Slice of the initialization methods that are applied in the statistical population of this review for Fuzzy JSSPs.
simulated annealing (SA), tabu search (TS), and articial immune system (AIS), are designed to improve the initial solutions of
Fuzzy JSSP, and complete the procedures of meta-heuristic algorithms. Fig. 15 shows the percentages of these state-of-theart improvement algorithms that are used for the statistical population of this study that are based on selected research papers as presented in Appendix A. As far as reviewed the literatures on this domain, there is not a literature on Fuzzy JSSP have
applied meta-heuristics such as electromagnetic-like mechanism (EM) and variable neighborhood search (VNS), and other
newly developed intelligent techniques such as neural-genetic approach, DNA computing, chemical-reaction optimization
(CRO), and gene expression programming (GEP). Table 3 shows the advantages and disadvantages of the improvement algorithms for JSSP.
Most studies have applied GA to solve Fuzzy JSSPs. Tsujimura et al. were the rst to apply GA in solving Fuzzy JSSP [84].
They considered the Fuzzy JSSP with the fuzzy processing time, and applied GA to minimize the fuzzy makespan. Ghrayeb
[1820] and Li et al. [47] also used GA in solving this type of Fuzzy JSSP. Sakawa and Mori [76] and Sakawa and Kubota [77]
were the rst to apply GA for Fuzzy JSSP with both fuzzy processing time and fuzzy due date. They incorporated the concept
of similarity among individuals into GA, while considering the fuzzy completion times in the three matrix m n as the fuzzy
decision variables. They also extended the G&T algorithm as the fuzzy schedule generator and maximized the agreement
index as the objective function. Their pursuits motivated a number of researchers such as [11,17,36,38,39,44,6870,73,90]
to apply GA on the Fuzzy JSSP. Lie [41] designed a co-evolutionary genetic algorithm (CGA) to solve fuzzy exible JSSP, with
the fuzzy processing time based on a new representation of chromosome, a new crossover operator, and a modied tournament selection in his algorithm. The failure of GA to intensify the search through the most promising regions of a neighborhood initiated the need to hybridize it with a local search algorithm or another meta-heuristic algorithm. Sakawa and Kubota
[76] hybridized GA with SA, Song et al. [79] combined GA and ACO, and Wang et al. [86] modied GA based on the immune
and entropy principles. Puente et al. [64] and Rodrguez et al. [70,72,74] enhanced GA by using a local search procedure for
Fuzzy JSSP with both fuzzy completion time and fuzzy due date. Petrovic and Fayad [63] hybridized a fuzzy shifting bottleneck procedure with GA to reduce the computational time that is individually taken by GA, and implemented this hybrid on
a real-world dataset. Wang et al. [85] designed the estimation of distribution algorithm (EDA) for fuzzy exible JSSP, which
employed explicit probability distributions in optimization. The only difference between the GA and EDA algorithms was
that the GA could reproduce a new population by crossover and mutation, whereas EDA could implicitly reproduce a new
population
Although particle swarm optimization (PSO) was proposed to solve continuous problems in optimization, this algorithm
was deemed as unsuitable for discrete problems such as Fuzzy JSSP. It was also ranked second among the meta-heuristics for
Fuzzy JSSP [35,40,43,45,46,58,97], (except for Lei [35], who solved Fuzzy JSSP with both fuzzy processing time and fuzzy due
date)]. Other bodies of literature used PSO to solve the Fuzzy JSSP with fuzzy processing time. Niu et al. [58] redened and
Fig. 15. Slice of the improvement algorithms that are proposed in the statistical population of this review for Fuzzy JSSPs.
392
Table 3
Advantages and disadvantages of meta-heuristic algorithms proposed for Fuzzy JSSPs.
Algorithm
GA
SA
TS
PSO
ACO
BCO
NN
AIS
Advantages
Chromosomes share information with each other
Resistant to be trapped in local optima
Easy in tuning the chromosomes and genetic
operators
Easy to understand and has no demand over complex knowledge of mathematics
Avoid becoming trapped in a local optimal
Search process can be controlled by the cooling
schedule
Ease of implementation
Has memory in its procedure, which is known as
the tabu list
Quick solution space search
Local search based
Particles update themselves with the internal
acceleration and velocity
Has memory and is suitable for global search
Particles give out the information to others
Positive feedback
Distributed computation avoids premature
convergence
Has colony memory
Fast convergence and high exibility
Fewer setting parameters
Memory of elite solutions by waggle dance
Quite simple to implement
Parallel processing capability
Powerful global exploration capability
Self-organizing and Simple training algorithm
Does not require effort to optimize any system
parameters
Disadvantages
Failure in intensifying the search through the most promising regions of a
neighborhood
Have no memory search based on random techniques
Require computational time more than other algorithm to reach the optimal
solution
Cannot reach good solutions of JSSP quickly
Memory-less technique
Difculty in dening an effective cooling schedule
modied the PSO by introducing genetic operators, such as crossover and mutation operators, to update the particles. Li et al.
[45,46] hybridized PSO with TS to improve the weaknesses of PSO in local search. A swarm-based neighborhood search
method was proposed by Lei and Guo [43] and Zheng et al. [97] to solve the single fuzzy exible JSSP and multi-objective
fuzzy exible JSSP, respectively. Two swaps, namely, an insertion and a tournament selection, were used to generate new
swarms in each generation.
The ant colony optimization (ACO) and bee colony optimization (BCO) algorithms have a limited application for Fuzzy
JSSP. Table 3 shows the advantages and the disadvantages of these algorithms. The implementation of these algorithms on
Fuzzy JSSPs is hybridized with other meta-heuristic algorithms. Song et al. [79,80] adopted ACO for Fuzzy JSSP with both
fuzzy processing time and fuzzy due date, and enhanced the adopted ACO by hybridizing it with GA and tabu search
(TS). TS is one of the powerful meta-heuristic algorithms to solve JSSPs, because most state-of-the-art algorithms for these
problems includes some sort of TS practicality. The main advantage of TS is that the memory in its procedure speeds up the
solution space search. Zheng and Li [98] used BCO for Fuzzy JSSP with both fuzzy processing time and fuzzy due date. They
also proposed a neighborhood structure based on the insertion operator to improve the search procedure.
The neural network (NN) is extendedly applied to solve Fuzzy JSSPs, although it is only applied in one previous literature.
Xie et al. [89] used the Hopeld neural network to solve the Fuzzy JSSP with fuzzy due date, and implemented the NN on
Ft06 with double fuzzy due dates. Fortemps [14] and Xu et al. [91] de-fuzzedFuzzy JSSP with fuzzy processing time based
on the mean value of a fuzzy number and then solved the related crisp JSSP by simulated annealing (SA) and articial immune system (AIS) respectively. SA cannot quickly achieve good solutions for JSSP because it is a generic.
technique that does not have memory. To block this disadvantage, Sakawa and Kubota [76] considered hybridizing SA
with GA. Hu et al. [27] used the differential evolution algorithm (DEA) to solve Fuzzy JSSPs with both fuzzy processing time
and fuzzy due date. DEA was proposed by [81] as a meta-heuristic continuous optimization algorithm. Similar to GA, DEA
contains three evolutionary operators, namely, selection, crossover, and mutation. Its signicant difference from GA is that
it utilizes information on distance and direction that is collected from the current population to guide the search process.
Unfortunately, the proposition of the meta-heuristic and intelligent algorithms remains in the infancy stage because only
a few studies have used TS, SA, and PSO, only one study has applied BCO, NN, AIS, and DEA, and the electromagnetic-like
mechanism (EM), variable neighborhood search (VNS), chemical-reaction optimization (CRO), rey algorithm (FA), and
imperialist competitive algorithm (ICA) have never been used.
393
(
e i d1 ; d2 )
D
i
i
di randdm ; dM
2
di randdi ; int1:3di
10
P
2
where dm ci 0:5qi and dM ci qi , ci m
j1 P ij were the sum of the most plausible values across all of its operations,
P
2
qij ri Prs was the sum of the most plausible values of all the other operations that require the same machine as operation
sj
(Oij) and qi maxj1;...;m qij . Song et al. [79,80] proposed a formula in Songs datasets to generate trapezoid fuzzy due dates
described as follows.
Fig. 16. Slice of the application of nine available groups of datasets in the statistical population of this review for Fuzzy JSSPs.
394
8 1
2
2
di di 0:2di
>
>
>
X 2
>
2
>
>
d 0:8 Pij
>
>
< i
j
1
2
3
4
e i d ; d ; d ; d )
D
X 2
i
i
i
i
3
>
>
d
P
> i
ij
>
>
j
>
>
>
: 4
2
2
di di 0:2di
11
Lins datasets have fuzzed a crisp instance with 9-job and 2-machine, Ft06, La01 and La06 based on the random technique,
to perform the Johnsons constructive algorithm [48,49]. Fortemps [14] built six-point fuzzy processing times for crisp Ft06
that was based on Ghrayebs datasets to evaluate SA. Wu et al. [87] considered the instance of Fortemps to examine the shifting bottleneck procedure. Ghrayeb [1820] also fuzzed the benchmark problems Ft06, La12, La13, and La14 to perform GA.
Ghrayebs datasets fuzzed Ft06, Ft10, Ft20, La01, La03, La05, La07, La09, Abz5, and Abz6 to perform the PSO that was
combined with genetic operators [58]. In Leis datasets for Fuzzy JSSP with preventive maintenance constraint [40],
Orb1-5, La20-22, and Abz5-6 were fuzzed to implement and evaluate the proposed swarm-based neighborhood search.
Rodriguez et al. [64,6874] used their random technique to fuzz Ft06, Ft10, Ft20, La21, La24, La25, La27, La29, La38, La40,
Abz7, Abz8, and Abz9, and applied them to perform an enhanced GA, memetic algorithm (MA) and local search procedure.
The hybridized ACO with GA and TS [79,80] were evaluated in Songs datasets, which fuzzed Ft10, La02, La19, La21, La24,
La25, La27, La29, La36, La37, La38, La39, and La40.
7. Conclusion
The classication of Fuzzy JSSPs, its constraints, objectives, and methodologies have been reviewed in this paper.
Although Fuzzy JSSPs are very important in the real-world JSSP situation, only a few studies in the literature have been published on this matter, which suggests that this kind of JSSP remains in the infancy stage.
Fifty four reviewed literatures are categorized based on three main fuzzy problems as in Fig. 2. This gure presents that
Fuzzy JSSP with both fuzzy processing time and fuzzy due date is the most frequent and Fuzzy JSSP with fuzzy due date is the
least frequent in the reviewed literatures. The causes of these frequents are summarized in Table 4 by presenting advantages
and disadvantages of three main Fuzzy JSSPs.
The methodologies for solving Fuzzy JSSPs are divided into three categories, namely, exact methods, heuristic approaches,
and meta-heuristic algorithms. Their advantages and disadvantages are discussed in Sections 4 and 5. Each algorithm is separately described based on the results and achievements as discussed in the previous literature. An outline of their advantages and disadvantages is presented in Table 5 to compare these methodologies. The table shows that meta-heuristic
algorithms are more effective than the other proposed methods, because they have successfully found the best-known solutions for the benchmark datasets on the real large scale of Fuzzy JSSP; nevertheless, they cannot guarantee the optimal
solutions.
Signicant attention on meta-heuristic algorithms has been utilized to review the algorithm. The meta-heuristics that are
applied for Fuzzy JSSP are divided into three steps, namely, pre-processing, initialization, and improvement. Pre-processing
consists of two tasks, namely, encoding the solution space and decoding the search space. Suitable choices of encoding
schemes and decoding algorithms are the prominent decisions, because they affect the exploration and exploitation procedures in the solution space. An effective choice of encoding scheme depends on the concepts and ideas that are applied to the
initialization and improvement of the algorithms. For instance, the machine-based encoding scheme is suitable for any algorithm that uses the shifting bottleneck procedure, the random key encoding scheme is useful for meta-heuristics with continuous features such as PSO, and the job pair relation-based encoding scheme is appropriate for neural networks. Each
encoding scheme does not have any limitations. They can be adapted and utilized for every algorithm, although not all of
them may be suitable.
Random initialization, application of priority rules to generate initial solutions, and heuristic initialization are the three
available approaches for the initialization. The high quality of the initial population speeds up the meta-heuristic algorithms
in solving Fuzzy JSSPs, and its cause is dependent on the solution space of Fuzzy JSSPs. Fig. 17 presents a sample of the solution space for Sakawas model as a Fuzzy JSSP. Most of the solutions in Sakawas model (e.g., the white region in Fig. 17) have
zero values, and a very small region from the solution space (e.g., the dark region in Fig. 17) has non-zero values that are
close to the high local optimal solutions. Therefore, in the execution of meta-heuristic algorithms, no substantial difference
exists among the solutions in the large part of the solution space, because they have zero values. Meta-heuristic algorithms
require a signicantly long time to exit from the white region of the solution space. The intelligent and heuristic initializations, which have generated an initial solution that is close to the local optimal, have successfully prepared initial solutions
with the non-zero objective function of Fuzzy JSSP. These initially produced solutions can be sensitively differentiated in the
execution of meta-heuristic algorithms. The non-existing heuristic initialization procedures, which are able to generate initial solutions that are close to the optimal solution, provide a clear and signicant gap for the Fuzzy JSSPs.
The intelligent techniques that are applied to improve the initial solution/population have been reviewed, and their
advantages and disadvantages are presented in Table 3. The diversity of the investigated algorithms is unsatisfactory, because 63% of the available literatures in Fuzzy JSSP have favorably considered GA to solve the problem. A number of different
395
Table 4
Advantages and disadvantages of three main Fuzzy JSSPs.
Advantages
Disadvantages
It does not have a standard objective function and proposed several objectives
It has fuzzed benchmark datasets that are generated by
random techniques
Table 5
Advantages and disadvantages of the proposed methodologies for Fuzzy JSSPs.
Advantages
Disadvantages
Exact methods
Heuristic approaches
Meta-heuristic algorithms
meta-heuristics have been tested on fuzzy JSSPs such as tabu search (TS), simulated annealing (SA), and particle swarm optimization (PSO). However, to date fuzzy JSSPs have not been investigated by such as electromagnetic-like mechanism (EM),
variable neighborhood search algorithms (VNS), gene expression programming (GEP), and chemical reaction optimization
(CRO). The advantages and disadvantages of improvement algorithms for Fuzzy JSSP (as presented in Table 3) and the application percentage of these algorithms in the previous literature (as shown in Fig. 15) highlight the effective features of these
algorithms to assist future research.
Appendix A. Summary on Fuzzy JSSP literatures based on bibliography, problem, methodology and dataset
In this appendix, the statistical population of the current review for Fuzzy JSSPs is categorized in Table 7 by bibliography,
problem, methodology and dataset. These literatures have been published between 1995 and 2013. The content of the tables
is ordered by the year of publication to represent the development of related techniques over the years. Figs. 2 and 1316
summarize the quantity of the problems, the representation schemes, initialization methods, improvement algorithms and
benchmark datasets applied for Fuzzy JSSPs, respectively.
396
Fig. 17. Sample of the solution space for Sakawas model of the Fuzzy JSSP.
In Table 7, the terms P1P3 indicate to Fuzzy JSSP with fuzzy due date, fuzzy JSSP with fuzzy processing time and
Fuzzy JSSP with both fuzzy due date and fuzzy processing time, respectively. The term P2 refers to Fuzzy Flexible JSSP
with fuzzy processing time. The term indicates that the corresponding properties were not presented in the paper.
The properties for the terms E1E9 and B1B9 are described in Table 6.
Appendix B. Mathematical formulations of Fuzzy JSSPs
The mathematical equations of eight fuzzy models are explained in this appendix. To explain the these fuzzy models of Fuzzy
JSSPs based on a mathematical formula, let J = {J1, J2, . . . , Jn} represent the set of jobs, M = {M1, M2, . . . , Mm} denote the set of
machines, and O fOij ji 1; 2; . . . ; n&j 1; 2; . . . ; mg be the set of operations to be scheduled. The Oij denotes the Jth operation
of the Ji. The operations of each job are interrelated by the predetermined orders (sequence of operations for jobs or SOJ). These
orders of machines are shown as a matrix of machines that the operations of jobs should be processed over their orders:
r
6.
SOJ 4 ..
..
.
3 2M
t
r
.. 7 6
..
6
5
.
4 .
k c
Mk
3
Mt
.. 7
7
. 5
Mc
..
.
12
Each row of SOJ corresponds to a given job and its elements represent a machine number, i.e. SOJij = s represents M s which
should be processed Oij. In other, if SOJij = s and SOJij+1 = h are assumed, where it means rst Oij should be processed on M s ,
and then the next operation of Ji can be processed on Mh, if Mh was idle. Furthermore, let Pij , Sij and Cij represent the xed
processing time of Oij as the constant of the problem, the starting time of Oij as the independent variable, and the completion
time of Oij as the dependent variable of JSSP, respectively. The rst fuzzy model, which is presented by Eqs. (13)(18), is Xies
model. The objective function of this fuzzy model maximizes the minimum satisfaction degree (Eq. (13)), which is illustrated
in Section 2 (see Eq. (2) and Fig. 4). Note that C im is the completion time of the last operation of J i which be used in Eq. (2) and
Fig. 4, are calculated in Eq. (14). The precedence constraints are represented by Eq. (15). Eqs. (16) and (17) denote to the
capacity constraints and also Eqs. (14) and (18) impose the release and due date constraints.
z MaxSDmin
Subject to : C im Sim Pim
13
14
15
SOJij SOJhr
17
Sij P 0
18
16
397
Table 6
The description of the terms E1E9 and B1B9.
Term
Property
Term
Property
E1
E2
E3
E4
E5
E6
E7
E8
E9
Operation-based representation
Job-based representation
Preference list-based representation
Job pair relation-based representation
priority rules-based representation
Disjunctive graph-based representation
Completion time-based representation
Machine-based representation
Random keys representation
B1
B2
B3
B4
B5
B6
B7
B8
B9
Sakawas datasets
Leis datasets for fuzzy exible JSSP
Xie datasets
Lins datasets
Ghrayebs datasets
Leis datasets for fuzzy JSSP
Rodriguezs datasets
Songs datasets
Other datasets
19
20
21
hr
hr
22
ij
SOJij SOJhr
23
DFSij P 0
24
Ghrayebs model tries to nd an optimal fuzzy schedule by minimizing both the total integral value and uncertainty of the
fuzzy makespan by Eq. (29) (see Eqs. (5) and (6)) that in Eq. (29), x = 0, 0.2, 0.4, 0.6, 0.8, 1, 2 and b e [0, 1]. Therefore, two
operations on the fuzzy quantity, sum and max, are required to generate fuzzy schedule and satisfy the constraints of the
problem. With respect to the fuzzy set theory was proposed by Zadeh [93], the fuzzy sum of two triangular fuzzy numbers
e ij P 1 ; P 2 ; P 3 is shown by the following formula:
(TFNs) e
S ij S1ij ; S2ij ; S3ij and P
ij
ij
ij
e
e ij
e ij S1 P1 ; S2 P2 ; S3 P 3 C
S ij P
ij
ij
ij
ij
ij
ij
25
e ij and C
e hr are l and l , respectively, then being consistent with the
Letting the membership functions of two TFNs C
eC ij
eC hr
e ij _ C
e hr through to the _ (fuzzy max) operation
extension principle of Zadeh [93], the membership function and le e , of C
C hr _ C ij
will be as follows:
26
Unfortunately however, since the fuzzy number obtained as a result of the _ (fuzzy max) operation through to the extension principle does not always become a TFN. In addition, to calculate a fuzzy schedule, the fuzzy interval is worked rather
than the membership function, and an approximation of the _ (fuzzy max) operation can be used. In Ghrayebs model, an
approximation of the fuzzy max was used that was proposed earlier by Sakawa & Kubota [76] and called Sakawas criterion
to approximate the fuzzy max in this dissertation. Thus the fuzzy max based on Sakawas criterion (s ) is formulated as
follows:
e ij _ C
e hr C 1 _ C 1 ; C 2 _ C 2 ; C 3 _ C 3 C
e
C
ij
hr
ij
hr
ij
hr
27
e ij S Cand
e
e hr S C
e
C
C
28
Therefore, Ghrayebs model can be formulated by Eqs. (29)(34), mathematically. The precedence constraints of
Ghrayebs model are represented by Eq. (31). Eqs. (32) and (33) denote to the capacity constraints of this model and Eqs.
(30) and (34) impose the release and due date constraints.
e max x U C
e max
z minTIV b C
e im e
e im
Subject to : C
S im P
29
e
e ij S e
S ij1
S ij P
8
Se
e
e
>
< S ij P ij S hr
or
>
:e
e hr S e
S ij
S hr P
31
SOJij SOJhr
S ij
0S e
33
30
32
34
398
Table 7
Summary of literatures based on bibliography, problems, methodologies and datasets.
References
Authors
Problem
Publisher
Methodologies
Preprocessing
Initialization
Improvement
Benchmark
datasets
Encoding
Decoder
The scheduling
problem is given by
the longest path
length in the
corresponding
digraph
Calculation of fuzzy
nishing times
Consistency
enforcing + Tree
Search + Lookahead analysis
B9
Random
Genetic
Algorithm
Priority rules
Branch and
Bound
Algorithm
Simulated
Annealing
B9-proposed 2
fuzzy datasets
with 6 6 and
20 5
B9-proposed 1
fuzzy dataset
with 8 4
B5-build a sixpoint fuzzy
processing time
for Ft06
Dubois et al.
[9]
Journal of intelligent
Manufacturing
Springer
P2
E6
Tsujimura
et al. [84]
Journal of Japan
Society for Fuzzy
Theory and Systems
P2
E1
Kuroda &
Wang [33]
International Journal
of Production
Economics Elsevier
IEEE Transactions on
Fuzzy System
P3
E3
P2
E6
International
Transactions in
Operational Research
Wiley Online Library
Computer and
Industrial Engineering
Elsevier
European Journal of
Operational Research
Elsevier
P3
E3
P3
E7
The possibility
distribution of
completion times
The scheduling
problem is given by
the longest path
length in the
corresponding
digraph
Order the jobs
according to the
shortest processing
time rule
P3
E7
Proceedings of
Industrial Engineering
Research
Thesis for PhD in New
Mexico State
University
Applied Soft
Computing Elsevier
ICCS 2001, LNCS 2074
Springer
P2
E1
Decoder procedure
proposed by [16]
Random
P2
E7
B2
Electronics and
Communications in
Japan Wiley Online
Library
Computer Integrated
Manufacturing
Systems Beijing
IEEE Transactions on
Fuzzy System
P3
E7
Johnsons
constructive
algorithm
Extended [21]
by random
selections
Genetic
Algorithm
B1
P3
E1
Gifer &Thompson
Algorithm (G&T)
Random
Genetic
Algorithm
B9
P2
E7
B2
Petrovic &
Fayad
[63]
P3
E6 & E3
Johnsons
constructive
algorithm
Priority rule:
early due date
Fayad &
Petrovic
[11]
Innovations in applied
articial intelligence
Springer
P3
E8 & E5
Priority rules
Li et al. [44]
Advances in Natural
Computation
Springer
IWINCA 2005, LNCS
3562 Springer
P3
E1
G&T
P3
E7
P1
E4
Extended [21]
by random
selections
Extended [21]
by random
selections
Fortemps
[14]
Sakawa &
Mori [75]
Sakawa &
Kubota
[76]
Ghrayeb [18]
Ghrayeb [20]
Ghrayeb [19]
Lin [48]
Sakawa &
Kubota
[77]
Geng & Zou
[17]
Lin [49]
Rodriguez
et al. [68]
Xie et al [89]
Random
Priority rules
Fuzzied model
of Moores [56]
Extended [21]
by random
selections
Extended [21]
by random
selections
Genetic
Algorithm
B1
Genetic
Algorithm +
Simulated
Annealing
Genetic
Algorithm
Shifting
Bottleneck +
Genetic
Algorithm
Fuzzy Genetic
Algorithm
Genetic
Algorithm
B5
B9: real-world
dataset collected
as Sherwood
Press
B9: fuzzed Becks
dataset [29]
Genetic
Algorithm
Neural Network
B3
399
Problem
Authors
Publisher
Computer Integrated
Manufacturing
Systems Beijing
Journal of Systems
Engineering Cnki
Lu et al. [51]
Methodologies
Preprocessing
Benchmark
datasets
Initialization
Improvement
Multi-Objective
Evolutionary
Algorithm
Improved
Genetic
Algorithm
Improved
Genetic
Algorithm
Genetic
Algorithm
B1
Shifting
Bottleneck
Procedure
Particle Swarm
Optimization
Genetic
Algorithm
B1
Genetic
Algorithm + Ant
Colony
Optimization
Ant Colony
Optimization +
Tabu Search
B8
Encoding
Decoder
P3
E5
G&T
Priority rules
P3
Quanyong &
Jianyung
[66]
Rodriguez et al.
[69]
Mechanical Science
and Technology Cnki
P3
P3
E7
Wu et al.
[87]
The International
Journal of Advanced
Manufacturing
Technology Springer
P2
E6
Extended [21]
by random
selections
P3
E7 & E5
Priority rules
P3
E1
Gifer &Thompson
Algorithm (G&T)
Random
P3
E1
Extended G&T
Random
P3
E1
Extended G&T
Random
P3
Random
Genetic
Algorithm
P2
E7
Random
Neural Network
P3
E1
G&T
Random
P2
E1
G&T
Random
P2
E6
Random
Genetic
Algorithm
Genetic
Algorithm +
Local Search
Local Search
P3
E7
P2
E1
Extended [21]
by similarity
between two
individuals
Random
B9: proposed a
numerical
problem with
34
B7: Ft06, La11,
La12, La13, La14
B1 & B7: Ft06,
La11, La12, La13,
La14
B1 & B7: Ft06,
Ft10, Ft20, La11,
La12, La13, La14,
La24, Abz7
B1
Rodriguez et al.
[74]
P2
E6
Random
Lei [36]
Computational
Intelligence in Flow
Shop and Job Shop
Scheduling Springer
Global Congress on
Intelligent Systems
IEEE
P3
E9
Gifer &Thompson
Algorithm (G&T)
Random
Genetic
Algorithm
P3
E1
Proposed a Decoder
Random
Genetic
Algorithm
Lei [35]
Xing et al.
[90]
TavakkoliMoghaddam
et al. [83]
Rodriguez et al.
[70]
Rodriguez et al.
[71]
Rodriguez et al.
[72]
Rodriguez et al.
[73]
Liu [50]
International
Conference on
Computational
Intelligence and
Security IEEE
Computational
Engineering in
Systems Applications
IEEE
Proceedings of the 6th
World Congress on
Intelligent Control and
Automation IEEE
Computer Integrated
Manufacturing
Systems Beijing
Journal of the
Operational Research
Society Palgrave
journal
CAEPIA 2007, LNAI
4788 Springer
Fuzzy Systems
Conference IEEE
Genetic
Algorithm
Particle Swarm
Optimization +
Genetic
Algorithm
Improved Local
Search
B1 & B7
B9
400
Table 7 (continued)
References
Problem
Methodologies
Authors
Publisher
Encoding
Decoder
Xu et al. [91]
Proceedings of the
rst ACM/SIGEVO
Summit on Genetic
and Evolutionary
Computation ACM
CAEPIA 2009, LNCS
5988 Springer
P2
E1
G&T
P2
E6
Lei [37]
International Journal
of Production
Research Taylor &
Francis
P2
E1 & E8
Lei [38]
Computers &
industrial engineering
Elsevier
The International
Journal of Advanced
Manufacturing
Technology Springer
The International
Journal of Advanced
Manufacturing
Technology Springer
P3
Puente et al.
[64]
Lei [39]
Hu et al. [27]
Lei [40]
Zheng & Li
[98]
Zheng et al.
[97]
International Journal
of Computer
Applications in
Technology
Inderscience
The International
Journal of Advanced
Manufacturing
Technology Springer
Li et al. [46]
Li et al. [47]
Swarm and
Evolutionary
Computation
Elsevier
The 24th Conference
on Control and
Decision IEEE
Applied Mathematics
and Computation
Elsevier
Preprocessing
Improvement
In the initial
step, the
antibodies are
randomly
generated
Random
Articial
Immune System
B9: 5 5
Local Search
Decoder procedure
proposed by [16]
Random
E9
G&T
Random
Efcient
decomposition
integration
Genetic
Algorithm
Genetic
Algorithm
P3
E9
G&T
Random
Genetic
Algorithm
P3
E1
G&T
Random
Differential
Evaluation
Algorithm
B1
P2
E1
Random
P3
E1
Proposed a decoder
Random
Particle Swarm
Optimization
Bee Colony
Optimization
B6: Orb1-5,
La20-22, Abz5-6
B9
P2
E1
Random
Swarm-based
Neighborhood
Search
P2
E1
Priority rules
and random
P2
E1
Priority Rules
and Random
Particle Swarm
Optimization +
Tabu Search
Particle Swarm
Optimization +
Tabu Search
B9: proposed 3
datasets 6 6 & 3
datasets 10 10
B9
P2
E1
G&T
Random
Genetic
Algorithm
P2
E1 & E8
Co-Evolutionary
Genetic
Algorithm
B9: 2 datasets
5 5 and 1
dataset 8 8
B2
Wang et al.
[85]
International Journal
of Production
Economics Elsevier
P2
E1 & E8
Wang et al.
[86]
International Journal
of Computer
Applications in
Technology
Inderscience
P2
E1 & E8
Initial
population P
with N
individuals are
randomly
generated
Initial
population P
with N
individuals are
randomly
generated
P2
International Journal
of Production
Research Taylor & Francis
E1 & E8
Lei [41]
Benchmark
datasets
Initialization
Estimation of
Distribution
Algorithm
Hybridized
Genetic
Algorithm with
Immune and
Entropy
principle
Swarm-based
Neighborhood
Search Algorithm
B1
401
Fig. 18. Seven possible positions for the triangular fuzzy processing time to meet the double fuzzy due date.
Rodriguezs model generates the fuzzy schedule by considering the fuzzy sum (Eq. (25)) and the approximation of fuzzy
max based on Sakawas criterion (Eq. (26) and (27)). Therefore the solution space of Rodriguezs model is similar to the solue max ) is considered as the objective function
tion space of Ghrayebs model. Minimizing the expected fuzzy makespan (min E C
in Rodriguezs model, and since the expected value of a TFN matches the neutral scalar replacement of a fuzzy interval, the
center of gravity of its mean value can also be obtained. The expected value of fuzzy makespan is calculated by Eq. (7). Thus,
Rodriguezs model is formulated by Eq. (39) as its objective function and Eqs. (30)(34) as its constraints.
Lei1s model is one type of Fuzzy JSSP with fuzzy processing time that tries to nd an optimal fuzzy schedule based on
minimizing the expected fuzzy makespan. The fuzzy sum (Eq. (25)) and an approximation of fuzzy max (L ) based on Leis
criterion have been considered to generate fuzzy schedules and satisfy the constraints of the problem. The approximation of
fuzzy max (L ) based on Leis criterion is formulated as described by the following equations:
1
e ij C ij 2C ij C ij ;
C1C
4
e ij C 2 ;
C2 C
ij
e ij C 3 C 1
C3C
ij
ij
35
402
Fig. 19. Position1position8 of AIi during which triangular fuzzy processing time will meet trapezoid fuzzy due date.
e hr C hr 2C hr C hr ; C 2 C
e hr C 2 ; C 3 C
e hr C 3 C 1
C1 C
hr
hr
hr
4
8
e ij < C 1 C
e hr ) C
e ij L C
e hr
>
C1C
>
< if
e ij C 1 C
e hr ; C 2 C
e ij < C 2 C
e hr ) C
e ij L C
e hr
elseif C 1 C
>
>
:
e
e
e
e
e
e hr ) C
e ij L C
e hr
elseif C 1 C ij C 1 C hr ; C 2 C ij C 2 C hr ; C 2 C ij < C 2 C
e ij L C
e hr ; then C
e ij _ C
e hr C
e hr ;
if C
e ij _ C
e hr C
e ij
else C
36
37
38
Therefore, Lei1s model can be formulated by Eqs. (39)(44), mathematically. The precedence constraints of Lei1s model
are represented by Eq. (41). Eqs. (42) and (43) denote to the capacity constraints of this model and Eqs. (40) and (44) impose
the release and due date constraints.
403
Fig. 20. Position9position16 of AIi during which triangular fuzzy processing time will meet trapezoid fuzzy due date.
e max
z minE C
e im e
e im
S im P
Subject to : C
39
e
e ij S e
S ij1
S ij P
8
Se
e
e
>
< S ij P ij S hr
or
>
:e
e hr S e
S ij
S hr P
41
SOJij SOJhr
S ij
0S e
43
40
42
44
404
Fig. 21. Position17position24 of AIi during which triangular fuzzy processing time will meet trapezoid fuzzy due date.
In Sakawas model, the fuzzy processing times (see Fig. 5) and fuzzy due date (see Fig. 4) considered for this model are
triangular fuzzy number and double fuzzy number, respectively. The fuzzy sum (Eq. (25)) and the approximation of fuzzy
max based on Sakawas criterion (Eqs. (27) and (28)) are applied to generate the fuzzy schedules. Maximizing the minimum
of the agreement index (the formulation of AIi is presented in Eq. (8) and Fig. 6) is the objective function of Sakawas model
and any fuzzy model in fuzzy JSSP with both of fuzzy processing time and fuzzy due date (Eq. 45). For Eq. (8), there are seven
e im \ D
e i , during which the fuzzy completion time will meet the fuzzy due date. As
possible positions to calculate the area C
shown in Fig. 18 there are occurrences where the fuzzy completion times will meet the fuzzy due date. So the mathematical
formulations of Sakawas model are presented as in Eqs. (46)(51).
405
Fig. 22. Position25 and position26 of AIi during which triangular fuzzy processing time will meet the trapezoid fuzzy due date.
z Maxmin AIi
16i6n
e im e
e im
S im P
Subject to : C
e
e ij S e
S ij1
S ij P
8
Se
e
e
>
S hr
S
P
ij
< ij
or
>
:e
e hr S e
S ij
S hr P
SOJij SOJhr
S ij
0S e
46
47
48
49
50
51
In Songs model, triangular fuzzy number (see Fig. 5) and trapezoid fuzzy number (see Fig. 7) are considered for the fuzzy
e ij ) and the due dates ( D
e i ), respectively. The fuzzy sum (Eq. (25)) and the approximation of fuzzy max
processing times ( P
based on Sakawas criterion (Eqs. (27) and (28)) are applied to generate the fuzzy schedules in this model. Also the objective
function of Songs model is maximizing the minimum of the agreement index (AIi), which is presented in Eq. (8) and Fig. 8. To
e im \ D
e i based on Eq. (8), during which the triangular fuzzy processing times will meet the trapezoid fuzzy
calculate area C
due dates, there are 26 possible positions (see Figs. 1922). The mathematical formulations of Songs model is similar the
formulations of Sakawas model and is represented by Eqs. (46)(51).
Lei2s model includes triangular fuzzy number (see Fig. 5) and double fuzzy number (see Fig. 4) for the fuzzy processing
e ij ) and the due dates ( D
e i ), respectively. The fuzzy sum (Eq. (25)) and the approximation of fuzzy max based on Leis
times ( P
criterion (Eqs. (35)(38)) are applied to generate the fuzzy schedules in this model. Maximizing the minimum of the agreement index (AIi), which is presented in Eq. (8) and Fig. 6, is the objective function of Lei2s model. Like Sakawas model, to
e im \ D
e i based on Eq. (8), there are seven possible positions (see Fig. 18). Therefore, Lei2s can be formulated
calculate area C
by Eqs. (52)(57). The precedence constraints of Lei2s model are represented by Eq. (54). Eqs. (55) and (56) denote to the
capacity constraints of this model and Eqs. (53) and (57) impose the release and due date constraints.
z Maxmin AIi
16i6n
e im e
e im
S im P
Subject to : C
e
e ij L e
S ij1
S ij P
8
Le
e
e
>
P
S hr
S
ij
< ij
or
>
:e
e hr L e
S ij
S hr P
SOJij SOJhr
S ij
0L e
52
53
54
55
56
57
References
[1] E. Balas, Machine sequencing via disjunctive graphs: an implicit enumeration algorithm, Oper. Res. 17 (1969) 941957.
[2] J.H. Blackstone, D.T. Phillips, L. GARY, A state-of-the-art survey of dispatching rules for manufacturing job shop operations, Int. J. Prod. Res. 20 (1982)
2745.
[3] C. Blum, A. Roli, Metaheuristics in combinatorial optimization: overview and conceptual comparison, ACM Comput. Surv. (CSUR) 35 (2003) 268308.
[4] K. Blbl, A hybrid shifting bottleneck-tabu search heuristic for the job shop total weighted tardiness problem, Comput. Oper. Res. 38 (2011) 967983.
[5] Y.B. Canbolat, E. Gundogar, Fuzzy priority rule for job shop scheduling, J. Intell. Manuf. 15 (2004) 527533.
[6] B. Chen, T.I. Matis, A exible dispatching rule for minimizing tardiness in job shop scheduling, Int. J. Prod. Econ. 141 (2013) 360365.
[7] R. Cheng, M. Gen, Y. Tsujimura, A tutorial survey of job-shop scheduling problems using genetic algorithmsI, Represent. Comput. Ind. Eng. 30 (1996)
983997.
406
[8] T.C. Chiang, L.C. Fu, Using dispatching rules for job shop scheduling with due date-based objectives, Int. J. Prod. Res. 45 (2007) 32453262.
[9] D. Dubois, H. Fargier, H. Prade, Fuzzy constraints in job-shop scheduling, J. Intell. Manuf. 6 (1995) 215234.
[10] V. Eswaramurthy, A. Tamilarasi, Hybridizing tabu search with ant colony optimization for solving job shop scheduling problems, Int. J. Adv. Manuf.
Technol. 40 (2009) 10041015.
[11] C. Fayad, S. Petrovic, A fuzzy genetic algorithm for real-world job shop scheduling, Innov. Appl. Artif. Intell. (2005) 7683.
[12] M.L. Fisher, Optimal solution of scheduling problems using Lagrange multipliers: Part I, Oper. Res. 21 (1973) 11141127.
[13] M. Florian, P. Trepant, G. McMahon, An implicit enumeration algorithm for the machine sequencing problem, Manage. Sci. 17 (1971) B-782B-792.
[14] P. Fortemps, Jobshop scheduling with imprecise durations: a fuzzy approach, IEEE Trans. Fuzzy Syst. 5 (1997) 557569.
[15] H.W. Ge, L. Sun, Y.C. Liang, F. Qian, An effective PSO and AIS-based hybrid intelligent algorithm for job-shop scheduling, IEEE Trans. Syst. Man Cybern.
Part A Syst. Hum. 38 (2008) 358368.
[16] M. Gen, R. Cheng, Genetic Algorithms and Engineering Optimization, Wiley-Interscience, 1999.
[17] Z. Geng, Y. Zou, Study on job shop fuzzy scheduling problem based on genetic algorithm, Comput. Integr. Manuf. Syst. Beijing 8 (2002) 616620.
[18] O. Ghrayeb, An efcient genetic algorithm for JSSP with fuzzy durations, in, 2000a.
[19] O. Ghrayeb, A bi-criteria optimization: minimizing the integral value and spread of fuzzy makspan of job shop scheduling problems, Appl. Soft Comput.
2 (3F) (2003) 197210.
[20] O.A. Ghrayeb, Solving Job-Shop Scheduling Problems with Fuzzy Durations Using Genetic Algorithms, New Mexico State University, 2000.
[21] B. Gifer, G.L. Thompson, Algorithms for solving production-scheduling problems, Oper. Res. 8 (1960) 487503.
[22] F. Glover, Heuristics for integer programming using surrogate constraints, Decis. Sci. 8 (1977) 156166.
[23] F. Glover, Future paths for integer programming and links to articial intelligence, Comput. Oper. Res. 13 (1986) 533549.
[24] H.H. Greenberg, A branch-bound solution to the general scheduling problem, Oper. Res. 16 (1968) 353361.
[25] C.W. Holsapple, V.S. Jacob, R. Pakath, J.S. Zaveri, A genetics-based hybrid scheduler for generating static schedules in exible manufacturing contexts,
IEEE Trans. Syst. Man Cybernet. 23 (1993) 953972.
[26] O. Holthaus, C. Rajendran, Efcient dispatching rules for scheduling in a job shop, Int. J. Prod. Econ. 48 (1997) 87105.
[27] Y. Hu, M. Yin, X. Li, A novel objective function for job-shop scheduling problem with fuzzy processing time and fuzzy due date using differential
evolution algorithm, Int. J. Adv. Manuf. Technol. 56 (2011) 11251138.
[28] T. Itoh, H. Ishii, Fuzzy due-date scheduling problem with fuzzy processing time, Int. Trans. Oper. Res. 6 (1999) 639647.
[29] J.C. Beck, M.S. Fox, Constraint-directed techniques for scheduling alternative activities, Artif. Intell. 121 (2000) 211250.
[30] A.S. Jain, S. Meeran, Job-shop scheduling using neural networks, Int. J. Prod. Res. 36 (1998) 12491272.
[31] A.M. Kuczapski, M.V. Micea, L.A. Maniu, V.I. Cretu, Efcient generation of near optimal initial populations to enhance genetic algorithms for job-shop
scheduling, Inf. Technol. Control 39 (2010) 3237.
[32] N. Kumar, G. Srinivasan, A genetic algorithm for job shop schedulinga case study, Comput. Ind. 31 (1996) 155160.
[33] M. Kuroda, Z. Wang, Fuzzy job shop scheduling, Int. J. Prod. Econ. 44 (1996) 4551.
[34] B.J. Lageweg, J.K. Lenstra, A.H.G.R. Kan, Job-shop scheduling by implicit enumeration, Manage. Sci. (1977) 441450.
[35] D. Lei, Pareto archive particle swarm optimization for multi-objective fuzzy job shop scheduling problems, Int. J. Adv. Manuf. Technol. 37 (2008) 157
165.
[36] D. Lei, Genetic algorithm for job shop scheduling under uncertainty, Comput. Intell. Flow Shop Job Shop Schedul. (2009) 191228.
[37] D. Lei, A genetic algorithm for exible job shop scheduling with fuzzy processing time, Int. J. Prod. Res. 48 (2010) 29953013.
[38] D. Lei, Fuzzy job shop scheduling problem with availability constraints, Comput. Ind. Eng. 58 (2010) 610617.
[39] D. Lei, Solving fuzzy job shop scheduling problems using random key genetic algorithm, Int. J. Adv. Manuf. Technol. 49 (2010) 253262.
[40] D. Lei, Scheduling fuzzy job shop with preventive maintenance through swarm-based neighborhood search, Int. J. Adv. Manuf. Technol. 54 (2011)
11211128.
[41] D. Lei, Co-evolutionary genetic algorithm for fuzzy exible job shop scheduling, Appl. Soft Comput. 12 (2012) 22372245.
[42] D. Lei, Z. Wu, Research on multi-objective fuzzy job shop scheduling, Comput. Integr. Manuf. Syst. Beijing 12 (2006) 174.
[43] D. Lei, X. Guo, Swarm-based neighbourhood search algorithm for fuzzy exible job shop scheduling, Int. J. Prod. Res. 50 (2012) 16391649.
[44] F. Li, Y. Zhu, C. Yin, X. Song, Fuzzy programming for multiobjective fuzzy job shop scheduling with alternative machines through genetic algorithms,
Adv. Nat. Comput. (2005) 429.
[45] J. Li, Y. Pan, A hybrid discrete particle swarm optimization algorithm for solving fuzzy job shop scheduling problem, Int. J. Adv. Manuf. Technol. (2012)
114.
[46] J. Li, Q.K. Pan, P. Suganthan, M. Tasgetiren, Solving fuzzy job-shop scheduling problem by a hybrid PSO algorithm, Swarm Evol. Comput. (2012) 275
282.
[47] J. Li, S. Xie, T. Sun, Y. Wang, H. Yang, Solving fuzzy job-shop scheduling problem by genetic algorithm, in: 2012 24th Chinese Control and Decision
Conference (CCDC), IEEE, 2012, pp. 32433247.
[48] F.-T. Lin, A job-shop scheduling problem with fuzzy processing times, in: ICCS 2001, LNCS 2074, Springer-Verlag, Berlin Heidelberg, 2001, pp. 409418.
[49] F. Lin, Fuzzy job-shop scheduling based on ranking level (k, 1) interval-valued fuzzy numbers, IEEE Trans. Fuzzy Syst. 10 (2002) 510522.
[50] J. Liu, Application of optimization genetic algorithm in fuzzy job shop scheduling problem, in: GCIS09, WRI Global Congress on Intelligent Systems,
2009, IEEE, 2009, pp. 436440.
[51] B. Lu, H. Chen, F. Gu, S. Yang, Research of earliness/tardiness problem in fuzzy job-shop scheduling, J. Syst. Eng. 6 (2006) 013.
[52] H. Lu, G.Q. Huang, H. Yang, Integrating order review/release and dispatching rules for assembly job shop scheduling using a simulation approach, Int. J.
Prod. Res. 49 (2011) 647669.
[53] A.S. Manne, Linear programming and sequential decisions, Manage. Sci. 6 (1960) 259267.
[54] G. McMahon, M. Florian, On scheduling with ready times and due dates to minimize maximum lateness, Oper. Res. 23 (1975) 475482.
[55] R.T. Moghaddam, M. Daneshmand-Mehr, A computer simulation model for job shop scheduling problems minimizing makespan, Comput. Ind. Eng. 48
(2005) 811823.
[56] J.M. Moore, An n job, one machine sequencing algorithm for minimizing the number of late jobs, Manage. Sci. 15 (1968) 102109.
[57] R. Nakano, T. Yamada, Conventional genetic algorithm for job shop problems, in: Proceedings of the Fourth International Conference on Genetic
Algorithms, San Mateo, 1991, pp. 474479.
[58] Q. Niu, B. Jiao, X. Gu, Particle swarm optimization combined with genetic operators for job shop scheduling problem with fuzzy processing time, Appl.
Math. Comput. 205 (2008) 148158.
[59] I.H. Osman, G. Laporte, Metaheuristics: a bibliography, Ann. Oper. Res. 63 (1996) 511623.
[60] S. Panwalkar, W. Iskander, A survey of scheduling rules, Oper. Res. 25 (1977) 4561.
[61] B.J. Park, H.R. Choi, H.S. Kim, A hybrid genetic algorithm for the job shop scheduling problems, Comput. Ind. Eng. 45 (2003) 597613.
[62] L.-J. Park, C.H. Park, Genetic algorithm for job shop scheduling problems based on two representational schemes, Electron. Lett. 31 (1995) 20512053.
[63] S. Petrovic, C. Fayad, A fuzzy shifting bottleneck hybridised with genetic algorithm for real-world job shop scheduling, in: Proceedings of Mini-EURO
Conference, Managing Uncertainty in Decision Support Models, Coimbra, Portugal, Citeseer, 2004, pp. 16.
[64] J. Puente, C. Vela, A. Hernndez-Arauzo, I. Gonzlez-Rodrguez, Improving local search for the fuzzy job shop using a lower bound, Curr. Top. Artif.
Intell. (2010) 222232.
[65] W. Qiao, B. Wang, J. Sun, Uncertain job shop scheduling problems solved by genetic algorithm, Comput. Integr. Manuf. Syst. Beijing 13 (2007) 2452.
[66] J. Quanyong, Z. Jianying, Study of fuzzy job shop scheduling problems with dualresource and multi-process routes, Mech. Sci. Technol. 12 (2006) 009.
407
[67] R. Ramkumar, A. Tamilarasi, T. Devi, A real time practical approach for multi objective job shop scheduling using fuzzy logic approach, J. Comput. Sci. 8
(2012) 606.
[68] I. Rodrguez, C.R. Vela, J. Puente, An evolutionary approach to designing and solving fuzzy job-shop problems, in: IWINAC 2005, LNCS 3562, SpringerVerlag, Berlin Heidelberg, 2005, pp. 7483.
[69] I. Rodrguez, C.R. Vela, J. Puente, Study of objective functions in fuzzy job-shop problem, in: ICAISC 2006, LNAI 4029, Springer-Verlag, Berlin Heidelberg,
2006, pp. 360369.
[70] I. Rodrguez, J. Puente, C.R. Vela, A multiobjective approach to fuzzy job shop problem using genetic algorithms, in: CAEPIA 2007, LNAI 4788, SpringerVerlag Berlin Heidelberg 2007, 2007a, pp. 8089.
[71] I. Rodrguez, J. Puente, C.R. Vela, A memetic approach to fuzzy job shop based on expectation model, in: Proceedings of Fuzzy-IEEE 2007, 2007.
[72] I. Rodrguez, C.R. Vela, J. Puente, R. Varela, A new local search for the job shop problem with uncertain durations, in, 2008, pp. 124131.
[73] I.G. Rodriguez, J. Puente, C.R. Vela, R. Varela, Semantics of schedules for the fuzzy job-shop problem, IEEE Trans. Syst. Man Cybernet. Part A: Syst.
Humans 38 (2008) 655666.
[74] I.G. Rodrguez, C.R. Vela, J. Puente, A. Hernndez-Arauzo, Improved local search for job shop scheduling with uncertain durations, in: Proc. of ICAPS,
2009, pp. 154161.
[75] M. Sakawa, T. Mori, An efcient genetic algorithm for job shop scheduling problems with fuzzy processing time and fuzzy due date, Comput. Ind. Eng.
36 (1999) 325341.
[76] M. Sakawa, R. Kubota, Fuzzy programming for multi-objective job shop scheduling with fuzzy processing time and fuzzy duedate through genetic
algorithms, Eur. J. Oper. Res. 120 (2000) 393407.
[77] M. Sakawa, R. Kubota, Two-objective fuzzy job shop scheduling through genetic algorithm, Electron. Commun. Jpn (Part III: Fundam. Electron. Sci.) 84
(2001) 6068.
[78] A. Shahzad, N. Mebarki, I. IRCCyN, Discovering dispatching rules for job shop scheduling problem through data mining, in: 8th International
Conference of Modeling and Simulation-MOSIM, 2010, pp. 1012.
[79] X. Song, Y. Zhu, C. Yin, F. Li, Study on the combination of genetic algorithms and ant Colony algorithms for solving fuzzy job shop scheduling problems,
in: IMACS Multiconference on Computational Engineering in Systems Applications(CESA), vol. 2, 2006, pp. 19041909.
[80] Y. Song, Y. Zhu, C. Yin, F. Li, A hybrid strategy based on ant colony and taboo search algorithms for fuzzy job shop scheduling, in: Proceedings of the 6th
World Congress on Intelligent Control and Automation, Dalian, China, 2006b, pp. 73627365.
[81] R. Storn, K. Price, Differential evolutiona simple and efcient heuristic for global optimization over continuous spaces, J. Global Optim. 11 (1997) 341
359.
[82] V. Subramaniam, T. Ramesh, G. Lee, Y. Wong, G. Hong, Job shop scheduling with dynamic fuzzy selection of dispatching rules, Int. J. Adv. Manuf.
Technol. 16 (2000) 759764.
[83] R. Tavakkoli-Moghaddam, N. Safaei, M. Kah, Accessing feasible space in a generalized job shop scheduling problem with the fuzzy processing times: a
fuzzy-neural approach, J. Oper. Res. Soc. 59 (2007) 431442.
[84] Y. Tsujimura, M. Gen, E. Kubota, Solving job-shop scheduling problem with fuzzy processing time using genetic algorithm, J. Jpn Soc. Fuzzy Theory Syst.
7 (1995) 10731083.
[85] S. Wang, L. Wang, Y. Xu, M. Liu, An effective estimation of distribution algorithm for the exible job-shop scheduling problem with fuzzy processing
time, Int. J. Prod. Res. (2013) 116.
[86] X. Wang, L. Gao, C. Zhang, X. Li, A multi-objective genetic algorithm for fuzzy exible job-shop scheduling problem, Int. J. Comput. Appl. Technol. 45
(2012) 115125.
[87] C.S. Wu, D.C. Li, T.I. Tsai, Applying the fuzzy ranking method to the shifting bottleneck procedure to solve scheduling problems of uncertainty, Int. J.
Adv. Manuf. Technol. 31 (2006) 98106.
[88] W. Xia, Z. Wu, A hybrid particle swarm optimization approach for the job-shop scheduling problem, Int. J. Adv. Manuf. Technol. 29 (2006) 360366.
[89] Y. Xie, J. Xie, J. Li, Fuzzy due dates job shop scheduling problem based on neural network, in: ISNN 2005, LNCS 3496, Springer-Verlag, Berlin Heidelberg
2005, 2005, pp. 782787.
[90] Y. Xing, Z. Wang, J. Sun, J. Meng, A multi-objective fuzzy genetic algorithm for job-shop scheduling problems, in: 2006 International Conference on
Computational Intelligence and Security, IEEE, 2006, pp. 398401.
[91] Z. Xu, X. Gu, B. Jiao, J. Gu, Research on Job Shop Scheduling Under Uncertainty, ACM, 2009.
[92] A. Yahyaoui, N. Fnaiech, F. Fnaiech, A suitable initialization procedure for speeding a neural network job-shop scheduling, IEEE Trans. Ind. Electron. 58
(2011) 10521060.
[93] L.A. Zadeh, Fuzzy sets, Inf. Control 8 (1965) 338353.
[94] C.Y. Zhang, P.G. Li, Y.Q. Rao, Z.L. Guan, A very fast TS/SA algorithm for the job shop scheduling problem, Comput. Oper. Res. 35 (2008) 282294.
[95] G. Zhang, L. Gao, Y. Shi, An effective genetic algorithm for the exible job-shop scheduling problem, Expert Syst. Appl. 38 (2011) 35633573.
[96] R. Zhang, C. Wu, A hybrid immune simulated annealing algorithm for the job shop scheduling problem, Appl. Soft Comput. 10 (2010) 7989.
[97] Y.-L. Zheng, Y.-X. Li, D.-M. Lei, Multi-objective swarm-based neighborhood search for fuzzy exible job shop scheduling, Int. J. Adv. Manuf. Technol. 60
(2012) 10631069.
[98] Y.L. Zheng, Y.X. Li, Articial bee colony algorithm for fuzzy job shop scheduling, Int. J. Comput. Appl. Technol. 44 (2012) 124129.