Job shop scheduling problem using genetic algorithm

Job Shop Scheduling problem(JSSP) using
Genetic Algorothm(GA)
(JSSP using GA)

Rakesh Kumar Chauhan
IMSEC GHAZIABAD
M.T.U, U.P
Noida, India
Rakeshchauhan49@gmail.com

Abstract—Job shop scheduling problem is one of the
most important problems in the combinatorial
optimization problems and it is applied to various fields
about engineering. Many works have been reported for
this problem using Genetic Algorithm (GA). The GA is one
of the most powerful optimization methods based on the
mechanics of natural evolution. In this paper, we propose a
new algorithm for the job shop scheduling using GA.In
this paper we used genetic algorithm(GA) with some

modifications to deal with problem of job shop
scheduling. In this paper, we generated an initial
population randomly including the result obtain by
some well known priority rules such as shortest
processing time and longest processing time. From
there, the population will go through the process of
reproduction, crossover and mutation to create a new
population for the next generation until some stopping
criteria defined were reached.
Keywords—Job-shop-Scheduling problem(JSSP),

Genetic Algorithm(GA), initial population,
crossover and mutation operation

I.

INTRODUCTION

Scheduling refers to the act of assigning resources to
tasks or assigning tasks to resources. The job-shop
scheduling problem is one of the most complicated and
typical. It aims to allocate m machines to perform n jobs in
order to optimize certain criterion.
The scheduling problem is a problem of deciding which job
in which order to be done with which machine, when
several jobs are processed with several machines. In
general, the scheduling problem is classified into several
types, if the problem has plural machines. The job shop
scheduling problem is one of the most difficult problems of
those types. The job shop scheduling problem has several
machines of which each machine has each function, and
each job is processed according to the specific order. Many

works have been reported for this problem using Genetic
Algorithm (GA).
Our intention in the research of Mahanim Omar, Adam
Baharum, Yahya Abu Hasan is to find out if the idea of
combining the CB neighbourhood and DG distance in
crossover and mutation is suitable when dealing with job-shop
scheduling problems so that the makespan value can be
minimized. Result has shown that if the solution converges too
quickly, it will stop at local optima. The modification has been
made so that it will get a solution at least not far from optima.
In this paper, the GA approach is used because it is one of
the approximate approaches that can handle a large searching
space, as in this integrated scheduling problem. There is no
Complex mathematical formula in developing GA and it can
also be understood easily. Furthermore, GA can produce a set
of acceptable solutions, compared to other search methods
such as simulated annealing, which gives only one solution.
II.

MODELING THE JOB-SHOP SCHEDULING PROBLEM

The job shop scheduling problem is a generalization of the
classical job shop problem. In the static job-shop scheduling
problem, finite jobs are to be processed by finite machines.
Each job consists of a predetermined sequence of task
operations, each of which needs to be processed without
preemption for a given period of time on a given machine.
Tasks of the same job cannot be processed concurrently and
each job must visit each machine exactly once. Each operation
cannot be commenced until the processing is completed, if the
precedent operation is still being processed. A schedule is an
assignment of operations to time slots on a machine. The
makespan is the maximum completion time of the jobs and the
objective of the JSSP is to find a schedule that minimizes the
makespan.
The job-shop scheduling problem (JSSP) can be described
as a set of n jobs denoted by J j where j =1,2,..., n which have
to be processed on a set of m machines denoted by Mk where
k =1,2,...,m. Operation of j th job on the k th machine will be
denoted by Ojk with the processing time Pjk . Once a machine
starts to process a job, no interruption is allowed. The time
required for all operations to complete their processes is called
makespan. In this paper, our intention is to minimize this
makespan value. every time when makespan is optimised, a
schedule can be described by the processing orders of
operations on the machines.

“Job-Shop Scheduling Problem (JSSP) is one of the well
known hardest combinatorial optimization problems. JSSP
being amongst the worst members of the class of NP-hard
problems” (Gary and Johnson, [1979]), there is still a lot of
room for improvement in the existing techniques. Because of
its large solution space JSSP is considered to be comparatively
one of the hardest problems to solve. “If there are n jobs and m
machines the number of theoretically possible solutions is
equal to (n!)m” (Noor [2007]). Among these solutions an
optimal solution, for a certain measure of performance, can be
found after checking all the possible alternatives. But the
checking of all the possible alternatives can only be possible
in small size problems. For example, a very simple problem of
5 jobs and 8 machines will give 4.3x1016 numbers of
alternatives. Even with a high performance computer, that can
evaluate one alternative per micro second, complete
enumeration of this problem to find out the optimal solution
would take more than 1000 years of continuous processing
(Hitomi [1996], Morshed [2006]).
Solution Techniques to Handle JSSP

A number of solution techniques to handle the JSSP
have been developed over the years. A broad
classification of the scheduling techniques is given in
Jain [1998]. Initially the techniques are divided into two
classes as approximation and optimization techniques.

A. Optimization Based Approaches
A lot of research work has been carried out, in this area,
in the last fifty years. Details are given in the following
sections.
I.

Johnson [1954] is one of the earliest
research works in this area. He developed a
heuristic based efficient method for finding an
optimal solution for the two and three stage
production scheduling problem, while considering
the setup time as well. Akers [1956] developed a
heuristic based graphical approach for solving the
production scheduling problem. Jackson [1956]
extended the Johnson’s rule and developed a
heuristic based approach for handling the job lot
scheduling. Hefetz and Adiri [1982] developed an
approach for two machines unit time Job-Shop
schedule length problem. These techniques are
applicable to very small problems and cannot
handle a big problem of more than 3 machines
efficiently.
II.

Figure: Solution approaches to JSSP
Optimization based techniques are further classified as
efficient techniques and enumerative techniques.
Enumerative approach has further two subclasses as
branch and bound algorithms and mathematical
optimization (mixed and linear integer programming)
based algorithms. On the other side approximation
techniques are initially classified as general algorithms
and tailored algorithms. Tailored algorithms are either
dispatching rules or heuristic based algorithms, whereas
general algorithms are classified as AI-based techniques
(ANN, GA and Expert Systems) and local search based
algorithms.

Efficient Techniques

Enumerative Techniques

Mathematical programming based approaches
have been extensively used to solve the JSSP. Balas
[1965] developed an addictive algorithm for solving the
linear programming model that had 0-1 variables. In his
[Balas (1969)] further work he developed a mixed
integer programming model for machine scheduling
using the disjunctive graphs. Mixed integer
programming models for the JSSP were also formulated
by Mann [1960], Giffler and Thompson [1960] and
Balas [1978]. Some of the researchers [Giffler and
Thompson (1960), Nemhauser and Wolsey (1988) and
Blazewicz et al (1991)] argued that mixed integer
programming had not been leading solution approaches
to the practical methods of solution. However, the
approach presented in Harjunkoski et al. [2000] is a
hybrid one, in which they formulated the JSSP as a
combination of mixed integer and constraint logic
programming.
B. Approximation Based Approaches:
Approximation based approaches offer a good
alternatives for solving the JSSP in terms of the quality of
solution and computational time.

I.

Tailored Algorithms

A problem of N jobs to be scheduled on M machines with
the objective of minimizing Makespan (time elapsed

between the start of the first operation and the
completion of the last operation), Cmax, can be handled
in two steps. In first step jobs are assigned to each
machine according to their processing requirements.
While, in second step those assigned jobs are sequenced
on each machine in such a way that Cmax is minimized.
The second step is normally handled by algorithms
known as Tailored Algorithm (Bedworth and Bailey [1987]).
II.

General Algorithms/ Artificial Intelligence:

General algorithms have been initially classified as
AIbased approaches and techniques based on local
search. According to Jain [1998] the AI-based
approaches, that mainly include GA and ANN, have
proved to be more efficient especially in the case of NPHard problems, where heuristic based solutions are
difficult to find.

Artificial Neural Networks (ANN)
A number of researchers applied ANN to the
JSSP. A neural network is a massively parallel

distributed processor made up of simple processing
units, which has a natural propensity for storing
experiential knowledge and making it available for
use. It resembles the brain in two respects:
1. Knowledge is acquired by the network from its
environment through a learning process.
2. Interneuron connection strengths, known as
synaptic weights, are used to store the acquired
knowledge.
III.

procedures with a variety of crossover and mutation schemes.
Representation is considered to be the first step in the
implementation of GA to a problem. According to Cheng et al
[1996] a total of nine different types of representation have
been used while applying GA to the JSSP. Details are as
under:
Representation based on:
Operations
Jobs
Preference list
Job-pair relation
Priority rule
Disjunctive graph
Completion time
Machines
Random keys
According to Cheng et al. [1996] the first five types
are termed as direct representations, whereas, the last four are
termed as indirect representations. In all the direct type of
representations a production schedule for a JSSP is directly
encoded as a chromosome, whereas, in case of indirect
representation a sequence of decisions related to scheduling a
system (for example dispatching rules) is encoded as a
chromosome. Morshad [2006] carried out a comprehensive
review of the different representation schemes used by
researchers. According to his survey, direct type
representation have been used by Nakano and Yamada [1991],
Yamada and Nakano [1992], Fang et al [1993], Gen et al
[1994], Norman and Bean [1995], Bierwirth et al [1996],
Masaru et al [2000],Wang and Zheng [2001], Zhou et al
[2001]; whereas, examples of indirect representation can be
found in; Falkenauer and Bouffoix [1991], Tamaki and
Nishikawa [1992], Della Corce et al [1995], Donrdoff and
Pesch [1995] Kobayashi et al [1995], Donrdoff and Pesch
[1995], Ghedjati [1999], Cai et al [2002].
After finalizing representation the next step is
selection of chromosomes that may take part in crossover and
mutation.

GENETIC ALGORITHM FOR JSSP
A. Chromosome Representation and Decoding

This kind of problem is known as NP-hard problem that can
be solved more efficiently with GA and other metaheuristic
methods. In this study, GA was selected to solve it. The major
mechanisms of GA are:
1. Creating the chromosome representation,
2. Creating the primary population,
3. Creating the adjustment function for fitness evaluation,
4. Defining the selection strategy,
5. Selecting genetic operators for creating a new generation,
6. Defining the parameter values
The four basic steps in the application of GA to a
problem are: representation, selection, crossover and mutation.
A number of research papers have been produced by different
researchers showing different representations and selection

The first step in constructing the GA is to define an
appropriate genetic representation (coding). A good
representation is crucial because it significantly affects all the
subsequent steps of the GA. Many representations for the JSP
problem have been develop.
In this study, two representations based on working
sequence and machine distribution are constructed. If the
number of the machine type t(t>1), the genes in each
chromosome will be divided into t parts in turn. Each part
represents one type of machine. Each operation can only be
assigned to the machines which can handle it. For example,
suppose a chromosome is given as [ 4221134233114234 ] in 4
job×4 machines problem. Here, 1 implies operation of job J1 ,
and 2 implies operation of job J2 , and 3 implies operation of

job J3 , and 4 implies operation of job J4 . Because there are
four operations in each job, it appears the four times in a
chromosome. Such as number 2 being repeated the forth in a
chromosome, it implies four operations of job J2 . The first
number 2 represents the first operation of job J2 which
processes on the machine 1. The second number 2 represents
the second operation of job J2 which processes on the
machine 2, and so on. The representation for such problem is
based on two-row structure, as following:

Figure . Chromosome Genes and Operations

Figure . Chromosome Genes and Machines

So the chromosome of machine 1 is 4 1 3 2, and the
chromosome of machine 2 is 2 3 1 4, and the chromosome of
machine 3 is 2 4 1 3, and the chromosome of machine 4 is 1 2
3 4.
B. Critical Block and DG distance
In job-shop scheduling problem, we also could find the
critical path. Critical path is defined as the longest paths taken
from the first operation processed until the last operation
leaves the workplace. All operations in this path are called
critical operations. Plus, the critical operations on the same
machine are called critical block.
We measure the distance between two schedules S1 and S2
by the number of differences in the processing orders of
operation on each machine (Mattfeld, 1996). By doing so, it is
just like summing the disjunctive arcs whose directions are
different between schedules S1 and S2 . This distance also
known as disjunctive graph (DG) distance.

C. Data Structures and Representation
Genetic algorithms process populations of strings. We
construct a population as an array of individuals where each
individual contains the phenotype, genotype (artificial
chromosome) and the fitness (objective function). In job-shop
scheduling problem we noted the job order on each machine
as phenotype, the machine schedule as genotype and makespan
value as fitness. The representation used in this paper is called
the permutation representation (Yamada and Nakano, 1997).
JSSP can be viewed as an ordering problem just like the
Travelling Salesman Problem (TSP). A schedule can be
represented by the set of permutations of jobs on each
machine which are called job sequence matrix. Figure shows
the job sequence matrix for 3× 3 problem. Rows will represent
the machines and columns represent the order of jobs.

Figure: A job sequence matrix for 3× 3 problem.
D. Initial population

Most of the research papers in this area prefer to
generate the initial population randomly. This technique
will allow the search process in a wide space. However,
this kind of technique will take a lot of time to get the
optimal value. In this paper, we chose to start the initial
population with a randomly generated schedules,
including some of the schedules obtain by the well
known priority rules such as the shortest processing time
and the longest processing time.
E. Crossover
Yamada and Nakano (1997) in most of their papers have
introduced plenty of techniques that could be use in solving
the job-shop problem. One of them is by making use of CB
neighbourhood and DG distance. The idea of this technique is
to evaluate a point x by the distance d(x, p2). Lets denote
parent1 and parent2 as p1 and p2 . First, set x = p1 . Then, we
generated the CB neighbourhood for x , N(x). For each
member, yi , in N(x) we calculated the distance between the
members and p2 to produce ( ) D yi , p2 . Then, we sort ( ) D
yi , p2 in ascending order. Starting from the first index in ( )
Dsort yi , p2 , we accepted yi with probability one if the fitness
value is less than the current fitness valueV(yi ) ≤ V(x).
Otherwise, we accepted it with probability 0.5. Starting from
p1 , we modified x step by step approaching p2 . After some
iteration, we will find that x will gradually loses p1 ’s
characteristics and started to inherit p2 ’s characteristics
although in a different ratios. We choose the child depending
on the less DG distance between the child and both its parents.
The algorithm for this procedure is shown in Algorithm 1.

Algorithm 1: Crossover
1. Let p1 and p2 be the parent solution.
2. Set x = p = q 1 .
3. Find CB Neighbourhood for x , N(x).
4. Do
a. For each member yi ∈ N(x), calculate the distance
between yi and p2 , ( ) D yi , p2 .
b. Sort the distance value in ascending order, ( )
Dsort yi , p2 .
c. Starting from i =1, do
i. Calculate the fitness value for yi , ( ) V yi .
ii. If V(yi ) ≤ V(x) accept yi with probability one,
and with probability 0.5 otherwise.
iii. If yi is not accepted, increase i by one.
Repeat i-iii until yi is accepted.
d. Set x = yi

e. If V(x)≤V(q) then set q = x .
Repeat 3-4 until number of iterations.
5. q is the child.
F. Mutation
Instead of using some random probability, we apply
mutation if the DG distance between parent1 and parent2 are
less than some predefined value. It is also defined based on the
same idea as crossover. However, we choose the child which
has the largest distance from the neighbourhood. The
algorithm for mutation is shown in Algorithm 2.
Algorithm 2: Mutation
1 Set x = p1.
2 Find Neighbourhood for x , N(x)
3 Do
a. For each member yi ∈ N(x), calculate the distance
between yi and p1 , ( ) D yi , p1 .
b. Sort the distance value in descending order, ( )
Dsort yi , p1 .
c. Starting from i = 1, do
i. Calculate the fitness value for yi , ( ) V yi .
ii. If V(yi ) ≤ V(x) accept yi with probability
one, and with probability 0.5 otherwise.
iii. If yi is not accepted, increase i by one.
Repeat i-iii until yi is accepted.
d. Set x = yi .
e. If V(x)≤V(q) then set q = x .
Repeat 2-3 until number of iteration.
q is the child.

4. If DG distance is large, we apply Algorithm 1 to p1 and p2 .
Generate child.
5. Apply neighbourhood search to child to find the fittest child
in the neighbourhood. Noted it as child’.
6. If the makespan for the child’ is less than the worst and not
equal to any member of population, replace the worst
individual with child’. If there is a member having the same
makespan value, replace the member with the child’.
7. Repeat 2-6 until some termination condition are satisfied.
IV.

METHODOLOGY

Having explained the approaches we provided in our
solution to solve the general job shop-scheduling
problem this section is dedicated to explain the steps of
the methodology. Our approach can be explained
according to the Figure 3.
This is the methodology by which our solution provides
the solution of the job shop scheduling problem.

G. Acceptance Criterion
The final and the most important step in the GA procedure
is to choose the individual to be replaced by child. It is widely
known that we always choose the fittest individual to
reproduce in the next iteration. In Yamada and Nakano (1997),
they did not consider or choose the child with the same fitness
value with other population members. However, by not even
considering that child, we may lose the good individual
without considering its abilities to be evaluated. So, in this
paper, after considering the worst individual in the population,
we also consider if the child has the same fitness value with
the member of population. Instead of dropping that child, we
replaced the old one with the child assuming that we have
given chance for the old individual to reproduce. Noted that
we could not take both of the individuals to avoid having
problem later, we might have problem falling in the local
optima. The algorithm for the whole procedure is shown in
Algorithm 3.
Algorithm 3: Genetic Algorithm
1. Initialize population: Randomly generated a set of 10
schedules including the schedules obtained by some priority
rules.
2. Randomly select two schedules, named them as p1 and p2 .
Calculate DG distance between p1 and p2 .
3. If DG distance is smaller from some predefined value,
apply Algorithm 2 to p1 . Generate child. Then go to step 5.

Figure: Methodology

V.

RESULTS AND DISCUSSIONS

Consider a 5 jobs and 5 machines problem with the
operation sequence and the processing time for each operation
have been determined in Table1 and Table2. We run the
programme for five times using the population size = 10,
number of iterations for mutation is 100 and crossover is 200.
The algorithm was terminated after 200 generations. From
the result in Table3, it can be shown that the combination of
critical block, DG distance and genetic algorithm could
provide a result as good as other methods. From Table 3, we
could see that the last job processed is job 5 on machine 4. So,
our makespan value for this problem is 34. The result also

gives us the job sequence for each machine to process, the
starting time and the finish time for each operation. For
example, on machine 1, we start to process job 3 at time 0 and
finished at 7. Then we process job 1, followed by job 4, job 5
and job 2.

We applied both types of initial population to the data. First
we used the combination of schedules we generated using the
priority rules and the randomly generated schedules as the
initial population. From the five runs, we find the optimum
before the generation exceeded 100. As for Figure 1.2, we
found the optimum value at generation 139. From the five
runs, we could conclude that if we used the randomly
generated schedules as the initial population, we will only find
the optimum value at generation larger than 100. However,
both results gave the same makespan value which is 34.

VI.

CONCLUSION & FUTURE WORK

The study on GA and job shop scheduling problem
provides a rich experience for the constrained
combinatorial optimization problems. Application of
genetic algorithm gives a good result most of the time.
Although GA takes plenty of time to provide a good
result, it provides a flexible framework for evolutionary
computation and it can handle varieties of objective
function and constraint.
In this paper GA for solving the job shop scheduling
to minimize the makespan. Two-row chromosome
structure is designed based on working procedure and
machine distribution. The relevant crossover and
mutation operation is also given.
Finally, The computational result shows that GA can
obtain better solution.
In future job-shop scheduling problem can have
the following aspect:
1. Operation-based representation
2. Machine-based representation
3. Preference list-based representation
4. Priority rule -based representation
5. Disjunctive graph-based representation

ACKNOWLEDGMENT
I take this opportunity to express my profound
gratitude and deep regards to my guide Mr. Vijay Singh
for his exemplary guidance, monitoring and constant
encouragement throughout the course of this repot. The
blessing, help and guidance given by him time to time
shall carry me a long way.

References
[1]
Ab. Rahman Ahmad and Faisal. RM Job Shop
Scheduling using GA. National Conference
Management Science and Operations Research 2003.
(2003).
[2] C. Bierwirth , D. C. Mattfeld and H. Kopfer. On
Permutation Representations for Scheduling Problems.
Parallel Problem Solving from Nature IV, SpringerVerlag, pp . 310-318 (1996).
[3] G. Mintsuo and C. Runwei.. GA and Engineering
Design, John Willey & Sons Inc, New York
USA.(2002).
[4]
J. Adams, E. Balas and D. Zawack The Shifting
Bottleneck Procedure for Job Shop Scheduling.
Management Science vol.34. (1988).

[5] J.F. Gonçalves, , Jorge José de Magalhães Mendes
and M. C. Resende. A Hybrid Genetic Algorithm for the
Job Shop Scheduling Problem..AT&T Labs Research
Technical Report TD- 5EAL6J, September 2002. (2002).
[6] M. Pinedo and X. Chao. Operation Scheduling with
Applications in Manufacturing and Services. McGrawHill International Editions. (1999).
[7] M. T. Jensen and T. K. Hansen. Robust Solutions to
Job Shop Problems. Proceedings of the 1999 Congress
on Evolutionary Computation, pages 1138-1144. (1999).
[8]
S. French. Sequencing and Scheduling : An
Introduction to the Mathematics of the Job Shop. John
Willey & Sons Inc, New York USA.(1982).
[9] T. Yamada and R. Nakano. Genetic Algorithms for
Job-shop Scheduling Problems. Proceedings of Modern
Heuristic for Decision Support. Pp. 67-81, UNICOM
Seminar, 18-19 March 1997,London. (1997)
[10] T. Yamada and R. Nakano. Scheduling by Genetic
Local Search with Multi-Step Crossover. The Fourth
International Conference on Parallel Problem Solving
from Nature, Berlin, Germany. (1996).
[11] T. Yamada and R. Nakano. A Genetic Algorithm
with Multi-Step Crossover for Job- Shop Scheduling
Problems. International Conference on Genetic
Algorithms in Engineering Systems:
InnovationsandApplication(GALESIA’95).(1995)
[12]
A Genetic Algorithm Approach for Solving a
Flexible
Job
Shop
Scheduling
Problem.
Sayedmohammadreza Vaghefinezhad1, Kuan Yew
Wong2…Department of Manufacturing & Industrial
Engineering, Faculty of Mechanical Engineering,
Universiti Teknologi Malaysia, 81310 UTM Johor
Bahru, Malaysia

Job shop scheduling problem using genetic algorithm

Recommended

More Related Content

What's hot (20)

Similar to Job shop scheduling problem using genetic algorithm (20)

Recently uploaded (20)

Job shop scheduling problem using genetic algorithm