An Improved MOEA/D for Multi-objective
Flexible Job Shop Scheduling with Release Time
Uncertainties
Xiaoning Shen, Yi Sun, Min Zhang
B-DAT, CICAEET, School of Information and Control
Nanjing University of Information Science and Technology
Nanjing, P.R. China
E-mail: sxnystsyt@sina.com.cn
Abstract—To capture the multi-objective and uncertain
nature of flexible job shop scheduling, a mathematical model
for the multi-objective flexible job shop scheduling problem
with release time uncertainties (MOFJSSP-RTU) is
constructed, where three objectives of make-span, tardiness,
and robustness are taken into account simultaneously under
various constraints. To solve MOFJSSP-RTU appropriately,
an improved multi-objective evolutionary algorithm based
on decomposition (IMOEA/D) is proposed for robust
scheduling. The novelty of our algorithm is that it employs a
new subproblem update strategy which utilizes the global
information, allows the elitists recorded in an archive to take
part in the child generation, and incorporates a repair-based
crossover operator and an adaptive differential evolution
(DE)-based mutation operator for variation, which helps
better balance the exploration and exploitation of the
algorithm. Experimental results on 4 problem instances
indicate that our IMOEA/D-based robust scheduling method
has a much better convergence performance than the state-
of-the-art multi-objective optimization evolutionary
algorithms (MOEAs), and it is also good at maintaining a
uniform distribution of solutions. Different trade-offs among
the three objectives are also analyzed.
Keywords—robust scheduling; multi-objective
optimization; decomposition; mathematical modelling;
I. I
NTRODUCTION
In flexible job shop scheduling problem (FJSSP), both
the machine assignment problem which selects an
alternative machine for each job operation, and the
sequencing problem which determines the best sequences
for processing jobs on each machine, should be addressed.
FJSSP is regarded as strongly NP-hard for its complexity
[1]. In most studies on FJSSP, a schedule is obtained by
just optimizing the efficiency objective such as tardiness,
assuming that all the problem information is deterministic.
However, in a realistic job shop, the manufacturing
environment often bears some uncertainties, such as job
release time delay that may stem from late deliveries of
subcontractors [2], or late arrivals of raw materials from a
supplier [3]. When facing uncertainties, the performance
of a previously optimal schedule may deteriorate. Thus, a
robust schedule which can optimize the robustness of the
schedule quality together with the efficiency is more
desirable for real-world applications, resulting in a multi-
objective optimization problem (MOP).
Multi-objective optimization evolutionary algorithms
(MOEAs) have been considered as well suited for MOPs
due to their capability to produce a set of non-dominated
solutions simultaneously in one run [4]. An MOEA based
on decomposition (MOEA/D) was first developed by [5],
and it presents a novel and general framework for dealing
with MOPs. It decomposes an MOP into a set of single-
objective subproblems, and these subproblems are solved
by a population based evolutionary algorithm in a
collaborative way. Due to its efficiency, various studies
on MOEA/D has been published in recent years [6] [7],
and it has been applied to a few real-world problems [8]
[9]. As far as we know, in the existing literature, although
MOEA/D has been used to solve permutation flow shop
scheduling [10] [11], it has not yet been investigated to
generate robust schedules in flexible job shops with
uncertainties, where the environment is much more
complex than the deterministic permutation flow shop.
In this paper, the multi-objective flexible job shop
scheduling problem with release time uncertainties
(MOFJSSP-RTU) is studied, where three objectives of
make-span, tardiness and robustness are considered
simultaneously. The problem is solved by an improved
MOEA/D (IMOEA/D) in a robust scheduling way. Our
IMOEA/D is different from the previous MOEA/D
versions in that: (1) a new subproblem update strategy
which utilizes the global information is developed; (2)
non-dominated solutions kept in an archive have a chance
to participate in the generation of child individuals; and (3)
an adaptive DE-based mutation operator which helps
better balance the exploration and exploitation of the
algorithm is designed. To validate the effectiveness of our
approach, some realistic flexible job shops are simulated
with two purposes: (1) comparing the convergence and
distribution performances in uncertain environments
generated by three MOEAs (IMOEA/D, MOEA/D-DE [7]
and NSGA-II [12]); and (2) investigating different trade-
offs among the three objectives.
The rest of this paper is organized as follows. Section
2 constructs the mathematical model of MOFJSSP-RTU.
Section 3 introduces the new IMOEA/D-based robust
scheduling approach. Experimental results are analyzed in
Section 4. Conclusions are drawn in Section 5.
This work is supported by the National Natural Science Foundation
of China (NSFC) under Grant No. 61502239 and No. 61503191, and
atural Science Foundation of Jiangsu Province of China under Grant
o. BK20150924 and No. BK20150933.
2950
978-1-5090-0623-6/16/$31.00
c
2016 IEEE