Co-evolutionary algorithm with MRF-based
decomposition mechanism for flexible scheduling
with uncertainty
Lu Sun, Student Member, IEEE Lin Lin *, Mitsuo Gen and Haojie Li, Member, IEEE
Abstract—Flexible scheduling problem plays an important
role in many real world applications as it can model many
practical issues, such the power grid scheduling, the logistics
scheduling, etc. Most researchers assume that the parameters
used in flexible scheduling problems are fixed. However, the
uncertain factors in real world problems cannot be ignored. In
this paper, we consider the uncertain processing times which are
modeled with the uniform distribution and design a hybrid co-
evolutionary algorithm with Markov random filed (MRF) based
decomposition mechanism for flexible scheduling problem with
uncertain processing times. All decision variables are decomposed
according to the learned network structure and the parameters
of MRF. The proposed MRF-based decomposition mechanism is
embedded into a cooperative co-evolutionary algorithm which is
named as CEA-MRF. Finally, the simulation results demonstrate
the effectiveness of our proposed algorithm compared with the
state-of-the-arts.
Index Terms—Cooperative co-evolution, Markov random filed,
stochastic flexible scheduling
I. INTRODUCTION
Flexible scheduling problem plays an important role in
many real world applications as it can model many practical
issues. Flexible scheduling can be described as a problem that
n jobs need to be processed on m machines with different
corresponding processing times. The objective is to minimize
the makepsan which is defined as the maximum comple-
tion time among all jobs through optimizing the schedule
with the suitable machine assignment [1]. Various researchers
concentrated on the approaches for optimizing the flexible
scheduling problems, and the literature pointed out that flexible
scheduling problem was proved as an NP-hard problem [2].
However, most studies only consider the ideal conditions in
which the processing times are fixed and estimated before
the schedule begins. In fact, the uncertain factors caused
by machine situation or the external environmental influence
cannot be ignored [3]. Therefore, in this paper, we focus on the
L. Sun is with School of Software, Dalian University of Technology, Dalian,
116620, China (e-mail:sunlu517@mail.dlut.edu.cn).
L. Lin is with International School of Information Science & Engineering,
Dalian University of Technology, Dalian, 116620, China; Fuzzy Logic Sys-
tems Institute, Japan; and Key Laboratory for Ubiquitous Network and Service
Software of Liaoning Province, Dalian University of Technology, China (e-
mail: lin@dlut.edu.cn). (Corresponding author: Lin Lin)
M. Gen is with Fuzzy Logic Systems Institute, Iizuka, Japan and Tokyo
University of Science, Tokyo, Japan (e-mail:gen@flsi.or.jp).
H. Li is with International School of Information Science & Engineering,
Dalian University of Technology, Dalian, 116620, China and Key Laboratory
for Ubiquitous Network and Service Software of Liaoning Province, Dalian
University of Technology, China (e-mail: hjli@dlut.edu.cn).
flexible scheduling problem with uncertain processing times,
and the uncertain processing times are modeled by the uniform
distribution.
Kundakc and Kulak studied a hybrid evolutionary algorithm
with local search strategy for solving stochastic job shop
scheduling problem (JSP) [4]. Hao et al. presented an effective
estimation of distribution algorithm with multiple objectives
for stochastic JSP [5]. Lei proposed a genetic algorithm with
multiple objectives to minimize the expected makespan of
stochastic JSP as well [6]. Zhang and Wu proposed an effective
artificial bee colony algorithm with novel update strategy for
stochastic JSP with the maximum lateness as objective [7].
Zhang et al. studied an effective particle swarm optimization
with two optimizing stages with the maximum lateness as
objective as well [8]. Azadeh et al. researched an effective op-
timization algorithms which based on artificial neural networks
(ANNs) to minimize the expected makespan of stochastic JSP
[9]. Horng et al. studied an novel optimization algorithm in
which two stages are collaborated with each other to minimize
the objective. [10].
Cooperative co-evolutionary algorithm (CEA) has attracted
researchers’ attention in recent years. CEA works through
searching the optimal solution among divided sub solution
spaces instead of in the whole solution space. Obviously, how
to divide various sub solution spaces is a key factor which
affects the effectiveness of CEA. The existing approaches
can be classified into two classes, i.e., random technique
based approaches and detected relationship based approaches.
For random technique based approaches, there exist three
types, i.e., single grouping, determined grouping and set-based
grouping. They stand for the strategy which are dividing
the decision variables into various groups in which each
group contains one decision variable, a determined number
of decision variables and a number which is from a given set
of decision variables, respectively. For detected relationship
basad approach, there exist delta grouping, K-means clustering
algorithm, cooperative co-evolution with variable interaction
learning, interdependence learning, fast interdependency i-
dentification, differential grouping, the extended differential
grouping [11]. They divide the decision variables according
to the detected relationship. However, compared with com-
binational optimization problems, they are more suitable for
the mathematical functions as they always work based on
difference theory [12].
Markov random filed (MRF) is an undirected graph which