International Journal of Informatics and Applied Mathematics

e-ISSN:2667-6990 Vol. 1, No. 1, 1-9

Hybrid Metaheuristic for Optimization Job-Shop

Scheduling Problem

Benhamza Karima and Zedadra Ouarda

LabSTIC, Computer Science Department, 8 Mai 1945 Guelma University, POB 401
Guelma, Algeria
(benhamza.karima, zedadra.ouarda)@univ-guelma.dz

Abstract. Real Job-shop scheduling problem is one of the most diffi-

cult NP-Combinatorial issues. Exact resolution methods cannot handle
large size cases. It is therefore necessary to use heuristic methods to solve
them within a reasonable time. There are a large number of metaheuris-
tic, which have the advantage of covering only part of the search space
to find an acceptable solution. In this work, Genetic Algorithm and Sim-
ulated Annealing are used to solve Job-shop scheduling problem. The
objective is to find the sequence of operations on the machines that will
minimize the total time required to complete the set of jobs, also known
as the ”Makespan”. Compared to traditional genetic algorithm, hybrid
approach yields significant improvement in solution quality.

Keywords: Scheduling Problem, Hybrid Metaheuristic, Optimization

2 Benhamza K. and Zedadra O.

1 Introduction
Job Shop Problem (JSP) is an important industrial activity in which jobs are
assigned to resources at specific times. Each job consists of a sequence of tasks,
which must be performed in a given order. Generating a good job scheduling
solution is particularly difficult in cases with large search space and with prece-
dence constraints between operations.
Lenstra and Rinnooy Kan [9] demonstrated that JSP is a difficult problem classi-
fied as NP-hard, so it cannot be exactly solved within a reasonable computation
time. To solve this problem, numerous approaches have been developed, incorpo-
rating exact and/or approximate methodologies. Exact methods such as branch
and bound and dynamic programming take significant computing time to get
the optimum solution.
Therefore, most presented JSP techniques have focused on the approximation
method, including Artificial Intelligence (AI) approaches. In recent years, meta-
heuristic algorithms have proven their flexibilities and efficiency for many com-
plex optimization problems. Several studies including Simulated Annealing, Tabu
Search, Genetic Algorithm, Ant Colony Optimization and Particle Swarm Op-
timization have been devoted to JSP scheduling problem [5].
Metaheuristic algorithms usually start their optimization process by generating
a population of random candidate solutions called individuals, then recombine
these initial individuals over a predefined number of iterations. The main dis-
tinction between different metaheuristic algorithms lies in the mechanism of re-
combination of individuals, which is often inspired by nature, biology or physics.
Genetic Algorithms (GA)[7] and Simulated Annealing (SA) [8] have become the
leading methodologies for search and optimization problems in high-dimensional
spaces. Genetic Algorithm (GA) is the best known and most widely used evolu-
tionary computation technique. There are essential steps in basic GA that can
be implemented differently according to the problem: population initialization,
fitness function computation, population selection, crossover and mutation op-
erators applied to a population of elements. These elements, called chromosomes
or individuals, represent possible solutions to the problem [4].
Although classical GAs are more resistant to premature convergence than other
search methods, GAs are not immune.
Hybrid methods offer advantages by avoiding this premature convergence and
allowing acceleration towards the AG algorithm. This research aims to propose
an efficient hybrid AG scheduling method to address this concern.
Moreover, Simulated Annealing (SA) has its origin in the fields of materials sci-
ence and physics. It is a search process most commonly used in optimization
problems. The objective function, similar to the energy of a material, is then
minimized, by introducing fictive temperature, which is a simple and control-
lable parameter of the algorithm. SA algorithm starts by generating an initial
solution and initializing the temperature parameter. Then, at each iteration, a
solution is randomly selected in the neighborhood of the current solution. SA
algorithm has the ability to find the optimal local result [10].
In this study, hybrid metaheuristic is used to optimize job-shop scheduling prob-
 Hybrid Metaheuristic for Optimization Job-Shop Scheduling Problem 3

lems. Two metaheuristic are considered: Genetic Algorithms (GAs) and Simu-
lated Annealing algorithm (SA). In JSP, the decision concerns how to sequence
operations on the machines, which will minimize the total time required to com-
plete the set of jobs, also known as the ”Makespan”.
The remainder of the paper is organized as follows: Section 2 presents the prob-
lem formulation. SA-GA-JSP Model is presented in Section 3, while experimental
results are described in Section 4. Finally, Section 4 summarizes the results of
this work and draws conclusions.

2 Problem Formulation

In JSP, a set of jobs is performed by a set of machines. Each job is formed by

a sequence of consecutive operations, and each operation requires exactly one
machine. Each machine is continuously available and can process one operation
at a time without interruption. The decision concerns how to sequence the op-
erations on the machines, which can minimize the makespan Cmax, i.e., the
maximum time to complete the final operation in the schedule operations. JSP
can be defined as follows [6]:
There is a set of n jobs: J = J1 , J2 , J3 , . . . , Jn and a set of m machines:
M = M1 , M2 , M3 , . . . , Mm . Each job Ji consists of a sequence of operations
Oi1 , Oi2 , Oi3 , . . . , Oini where ni is the number of operations that Ji comprises.
Each operation Oij (i = 1, 2, , n; j = 1, 2, , ni ) must be processed by one machine
on a given set of machines Mij ⊆ M. The respected hypotheses are summarized
as follows:

− Jobs are independent, each machine can handle only one job at a time.
− All jobs and machines are available at time zero simultaneously, operations of
one job cannot be processed simultaneously.
− After a job is processed on a machine, it is immediately transported to the next
machine on its process, and the transmission time is assumed to be negligible.

3 SA-GA-JSP Model

In this work, sequential hybrid method based on GA and SA is exposed. Thus,

SA is applied after the GAs to exploit the result of the previous exploration
of the GA search space. The process adopted is to wait for the stabilization of
the fitness in the GAs population. The best individuals found are the starting
solutions of the SA algorithm. Figure 1 presents the general hybrid SA-GA-JSP
model. The following section describes the main steps of SA-GA-JSP.

3.1 GA-JSP

In the GA, a suitable encoding (or representation) for the problem must first
be devised. A fitness function is also required, which assigns a merit figure to
4 Benhamza K. and Zedadra O.

Fig. 1: SA-GA-JSP model

each encoded solution. The first population of solutions is generated randomly.

During execution, parents must be selected for reproduction, and recombined to
generate offspring (population of new solutions). The process is repeated until a
good solution is provided. The stop criterion is generally based on the number
of generations. These stages are detailed as follows:

a. Encoding representation: First task in GA-JSP is the solution repre-

sentation as chromosome. An operation-based representation is used. In this
formulation, each job number occurs m times in the permutation, i.e. as often
as there are operations associated with this job.
For example, with 4 jobs x 3 machines problem (each job consists of three op-
erations), a chromosome is given as [1 2 1 2 3 1 1 2 2 3 3 3] related to jobs
representation [1 1 2 2 1 3 4 3 4 2 3 4] and associated operations-based repre-
sentation[1 2 1 2 3 1 1 2 2 3 3 3].

b. Fitness function: GA assesses the solutions based on the fitness function.

The better fitness an individual has, the higher the chance of being chosen for
the next generation. In general, the fitness is relative to the objective function. In
this paper, the minimal makespan represents a good objective measure function
U(k). Then, the fitness function for solution k is as follows:

1
F (k) = (1)
U (k)

c. GA operation phases: In GA operation phase, an initial population is

yielded randomly. Therefore, GA operates to produce new population by using
basic genetic operations, i.e. selection, crossover and mutation.
 Hybrid Metaheuristic for Optimization Job-Shop Scheduling Problem 5

Start

Generate a random initial population

Evaluate individual's fitness

Select best individuals for reproduction

Crossover and mutation

Evaluate offspring's fitness

Select best offsprings for substitution

No
Is stoping creteria
satisfied

Yes

Best solution

Fig. 2: GA-JSP Flow Chart

1. Selection operation: In GA, parent chromosomes are selected with a probabil-

ity related to their fitness. The standard procedure is a fitness-proportional
selection (or roulette wheel selection) of individuals to produce the offspring.
A roulette wheel model is established to represent the survival probabilities
for all the individuals in the population [7].
The probability P(i) of selecting the i-th individual is given by:

F (i)
P (i) = Pps (2)
k=1 F (k)
6 Benhamza K. and Zedadra O.

This means that each individual is interpreted as a segment of the wheel so

that the size of this segment is proportional to the fitness. In general, by
means of two chosen parents, one or two offspring can be generated.

2. Crossover operation: The crossover is the main operator applied in genetic

algorithms. By crossover, the genetic structure of two selected individuals

AG_Best Solution

Evaluate Initial solution

Generate a new solution

Evaluate a new solution

New solution better Calcul.

than old solution P= exp(- E/T)

No p<Random N0

Replace old solution with Yes

new solution

Decrement temperature

No Stopping
criteria?

Yes

Optimized solution

Fig. 3: SA-GA-JSP Flow Chart

of the current population is mixed. In this work, the idea for scheduling
problems resulting in precedence preservative crossover operator (PPX) is
integrated [3]. The principle is summarized as follows: A vector of equal
length defines the order in which the operations are successively drawn from
parent 1 and parent 2. First, offspring is initialized empty. Then, an opera-
tion is selected; it is deleted in both parents and appended to the offspring.
The leftmost entry of the binary vector is also deleted. This procedure is
 Hybrid Metaheuristic for Optimization Job-Shop Scheduling Problem 7

repeated until the parent lists are empty and the offspring list contains all
operations.

3. Mutation operation: Mutation is used only to produce small perturbations

on chromosomes in order to maintain the diversity of population. For the
operation sequence, two positions were selected and exchanged with each
other. The mutation rate was applied with a probability of 0.1.

3.2 SA-GA-JSP

The best individual found in Step 1 was used as first solution. Initial temperature
was set, then a loop was started until the threshold was reached. A neighbor was
selected by making a small change of the current solution. Using the Metropolis
rules, the transition to the neighbor was accepted or not. Finally, the temperature
was slowly decreased until the thermodynamics equilibrium was reached. The
solution converges as shown in Figure 3.

4 Experimental Results

A number of benchmark problems for JSP have been formulated in literature

to compare the performance of new scheduling algorithms [2]. For the purpose

2690

2490
GA-JSP
SA-GA-JSP

2290

2090
Makespan

1890

1690

1490

1290
0 1000 2000 3000 4000 5000 6000
Itérations

Fig. 4: The makespans evolutions for JSP example (25 jobs and 08 machines)
8 Benhamza K. and Zedadra O.

of this study, 30 instances were selected from OR-Library [1] as benchmarks

to test SA-GA-JSP model. 180 chromosomes were used as initial population
and the crossover rate was set at 0.6. Temperature T was initialized to 3000
to measure the quality of proposed system. Implementation was successfully
completed to examine the convergence and effectiveness of the hybrid method.
Figure 4 shows that the hybrid method reaches the optimal solution faster than
GA-JSP method and produce less makespan value. Empirical results show that
hybrid SA-GA-JSP method is effective enough to obtain a better solution than
the genetic algorithm applied individually. SA-GA-JSP model converges towards
the optimal solution for most instances of the benchmark problems.

5 Conclusion

Genetic Algorithms (GA) and Simulated Annealing (SA) represent powerful

combinatorial optimization methods. In this work, an effective combination of
GA and SA was developed to solve the job shop scheduling problem (JSP). This
model has the advantage of avoiding premature convergence and allowing ac-
celeration to AG algorithm. GA is suitable for traversing large search spaces in
compared to other optimization methods. SA, alternatively attempts to avoid
being trapped in a local optimum.
This hybrid metahaeuristic gives an excellent trade-off between solution quality,
computing time and flexibility to take into account specific constraints in real
JSP scheduling situations. Indeed, GA is effective for good exploration and SA
is the best heuristic for task exploitation.

References

1. J. E. BEASLEY. Or-library. http://people.brunel.ac.uk/mastjjb/jeb/info.html.

2. Dennis Behnke and Martin Josef Geiger. Test instances for the flexible job shop
scheduling problem with work centers, 2012.
3. Christian Bierwirth, Dirk C Mattfeld, and Herbert Kopfer. On permutation repre-
sentations for scheduling problems. In International Conference on Parallel Prob-
lem Solving from Nature, pages 310–318. Springer, 1996.
4. Ilhem Boussaı̈D, Julien Lepagnot, and Patrick Siarry. A survey on optimization
metaheuristics. Information sciences, 237:82–117, 2013.
5. Banu Çaliş and Serol Bulkan. A research survey: review of ai solution strategies of
job shop scheduling problem. Journal of Intelligent Manufacturing, 26(5):961–973,
2015.
6. Liang Gao, CY Peng, C Zhou, and PG Li. Solving flexible job shop scheduling
problem using general particle swarm optimization. In Proceedings of the 36th CIE
conference on computers & industrial engineering, pages 3018–3027, 2006.
7. David E Goldberg. Genetic algorithms in search. Optimization, and Machine-
Learning, 1989.
8. Scott Kirkpatrick. Optimization by simulated annealing: Quantitative studies.
Journal of statistical physics, 34(5-6):975–986, 1984.
 Hybrid Metaheuristic for Optimization Job-Shop Scheduling Problem 9

9. Jan K Lenstra and AHG Rinnooy Kan. Computational complexity of discrete

optimization problems. In Annals of Discrete Mathematics, volume 4, pages 121–
140. Elsevier, 1979.
10. Peter JM Van Laarhoven and Emile HL Aarts. Simulated annealing. In Simulated
annealing: Theory and applications, pages 7–15. Springer, 1987.