A Memetic Algorithm for Multi-objective
Optimization Problems with Interval Parameters
Dunwei Gong
School of Information and
Electrical Engineering
China University of Mining and Technology
Xuzhou, China
Email: dwgong@vip.163.com
Zhuang Miao
School of Information and
Electrical Engineering
China University of Mining and Technology
Xuzhou, China
Email: 1037407168@qq.com
Jing Sun
School of Sciences
Huaihai Institute of Technology
Lianyungang, China
Email: sunj@hhit.edu.cn
Abstract—Multi-objective optimization problems with interval
parameters (IMOPs) are ubiquitous in real-world applications.
The existing evolutionary algorithms for IMOPs (IMOEAs)
require a large amount of function evaluations to generate an
approximate Pareto front which is well converged and evenly
distributed, and the generated front has uncertainties to a large
extent. In this paper, a local search is embedded into an existing
IMOEA, and a memetic algorithm for IMOPs is developed. The
existing IMOEA is first employed to search the entire search
space, and then the rate of changes of hypervolume is utilized
to design an activation mechanism to specify when to conduct
the local search. Finally, an initial population of the local search
is created by taking the individuals with a large contribution to
hypervolume and a small imprecision as the center, and the local
search is implemented by taking the contribution to hypervolume
as its fitness function. The proposed algorithm is applied to six
benchmark IMOPs and an uncertain optimization problem of
solar desalination, and compared with a typical IMOEA without
the local search. The empirical results indicate the effectiveness
of the proposed algorithm.
I. INTRODUCTION
In real-world situations, a variety of practical problems
can be formalized to optimization problems. Because the
optimization problems usually contain more than one objective
function, and these objectives conflict each other, this kind
of the problems is termed as multi-objective optimization
problems, e.g., device of radiator [1], design of electric power
distribution system [2], and flexible job-shop scheduling [3],
etc. If the objective function(s) or (and) constraint(s) of the
optimization problems contain(s) some uncertainties, they are
called uncertain optimization problems, where IMOPs are not
only ubiquitous, but also extremely important. For this kind
of optimization problems, the parameters of their objective
function(s) or (and) constraint(s) are usually interval numbers,
whose upper and lower bounds or midpoint and radius of
the intervals are known in advance. As a consequence, the
optimization problem with interval parameters have been ap-
plied in various practical problems, e.g., allocation of water
resources [4] and portfolio [5], etc. Without loss of generality,
the following IMOPs are focused on in this paper:
min
f (x, c)=(f
1
(x, c
1
) ,f
2
(x, c
2
) , ··· ,f
m
(x, c
m
))
s.t.
x ∈ S ⊆ R
n
,
c
i
=(c
i1
,c
i2
, ··· ,c
il
)
T
,c
ik
=[c
ik
, c
ik
] ,
i =1, 2, ··· ,m,k =1, 2, ··· ,l,
(1)
where x is an n-dimensional decision variable; S is the deci-
sion space of x; f
i
(x, c
i
)(i =1, 2, ··· ,m) is the ith objective
function with interval parameters; c
i
is a parameter described
with an interval vector; c
ik
is the kth component with c
ik
and
c
ik
being its lower and upper bounds, respectively. Each ob-
jective function of problem (1) is an interval due to its interval
parameters, and denoted as f
i
(x, c
i
)
=
f
i
(x, c
i
) , f
i
(x, c
i
)
.
Although existing evolutionary optimization methods [6-11]
provide a variety of feasible ways to solve the above problems,
a large amount of objective evaluations are still required to
effectively approach to the true Pareto front. In addition, these
methods aim to generate an approximate Pareto front which
is well converged and evenly distributed, and neglect to deal
with uncertainties involved in the objectives, which directly
affect the stability of the system.
Memetic algorithms (MAs) are hybrid optimization methods
which embed a local search mechanism into the process of
evolutionary optimization [12]. These methods can speed up
the convergence and obtain a high-performance approximate
Pareto front. So they attract many researchers’ interest [12-
16]. In view of this, a memetic algorithm for solving multi-
objective optimization problems with interval parameters (I-
MOMA) is proposed.
The main contributions of this paper are as follows: (1) a
memetic algorithm is adopted to solve IMOPs, which not only
improves the algorithm’s performance, but also reduces the
imprecision of the final approximate Pareto front; (2) the lower
bound of threshold for an activation mechanism of a local
search is set to prevent the algorithm from frequently activating
a local search, and thus avoiding the waste of computing
resources; (3) the imprecision of an individual is taken into
account in the initial and the offspring populations of each lo-
cal search, therefore the evolutionary individuals with smaller
imprecisions are bound to be produced by each local search;
1674
978-1-5090-0623-6/16/$31.00
c
2016 IEEE