Decision SupportA multiobjective evolutionary algorithm for approximating the efficient set
Introduction
For multiobjective decision making (MODM) problems (see, e.g., Hanne, 2001a, Steuer, 1986, Vincke, 1992, Zeleny, 1982), a significant number of algorithms based on evolutionary approaches has been proposed during the last 15 years. Today, there are various survey articles of this research field available (see Fonseca and Fleming, 1995, Horn, 1997, Tamaki et al., 1996), specialized international conferences on evolutionary multi-criterion optimization take place (see the proceedings edited by Zitzler et al., 2001, Fonseca et al., 2003), and comprehensive monographs have been published (see Coello Coello et al., 2002, Deb, 2001). Theoretical results on evolutionary algorithms for multiobjective optimization such as, for instance, approximation proofs are, however, scarcely available. In this paper we introduce a new framework for evolutionary multiobjective algorithms which allows for an analysis of approximation. Approximation, in that sense, means that, providing that sufficient computation time and memory is available, the algorithm is capable of reaching with an arbitrary exactness those alternatives which constitute in a mathematical sense the solution set which is usually denoted as the set of efficient or Pareto-optimal alternatives. The subsequent paper is organized as follows: In the next section some notation on multiobjective optimization problems is specified. In Section 3, an algorithm (or: algorithmic framework) capable of approximating the efficient set is introduced. In Section 4, further theoretical concepts for an analysis of approximation and additional assumptions are introduced and used for a corresponding proof. In Section 5, some conclusions are given.
Section snippets
Some notations
Originally, evolutionary algorithms (EAs) have been developed for scalar, or ordinary, optimization problems (see, e.g., Bäck et al., 1991, Bäck et al., 1997), i.e. problems with a mathematical formulation as follows:with A ⊆ Rn called the feasible set and f : Rn → R being the objective function to be maximized. Instead of maximization a minimization can be assumed as well. The feasible set is usually defined by constraint functions,Elements a ∈ A are usually
General framework
In the following, we sketch a new framework for multiobjective evolutionary algorithms suitable for approximating the efficient set of a multiobjective optimization (MOO) problem (2.4), (2.5).
The basic idea of EAs is that from a set (population) of intermediary solutions (feasible alternatives) a subsequent set of solutions is generated by imitating concepts of natural evolution such as mutation, recombination, and selection. From a set of ‘parents’, Mt, in generation t ∈ N, a set of ‘offspring’,
Theoretical results on approximation
In Hanne (1999) we have introduced a new concept of ε-efficient solutions used for analyzing the approximation features of our algorithm. This concept is based on ε-neighborhoods and does not require scalarization as some other definitions of ε-efficient solutions do (cf. Helbig and Pateva, 1994). Quite frequently (see, e.g. Laumanns et al., 2002, Ehrgott and Gandibleux, 2004) ε-efficiency is defined component-wise for the objective functions which can be considered as using a specific metrics
Conclusions
In this article we have presented a new framework for multiobjective evolutionary algorithms for approximating the efficient set, denoted as AMOEA. Roughly speaking, this is an extension of our earlier MOEA with a variable and usually increasing population size such that enough alternatives can be generated and kept in population for representing the efficient set with a given exactness ε. The general framework of the algorithm is presented and implementation details of the single steps are
References (39)
Approximation algorithms in combinatorial multicriteria optimization problems
International Transactions in Operational Research
(2000)On efficient sets in vector maximum problems—a brief survey
European Journal of Operations Research
(1986)On the convergence of multiobjective evolutionary algorithms
European Journal of Operational Research
(1999)- et al.
A survey of evolution strategies
The Theory of Evolution Strategies
(2001)- et al.
Evolutionary Algorithms for Solving Multi-Objective Problems
(2002) Multi-Objective Optimization Using Evolutionary Algorithms
(2001)- et al.
Approximative solution methods by for multiobjective combinatorial optimization
TOP
(2004) - et al.
A rapidly convergent descent method for minimization
Computer Journal
(1963)
Genetic algorithms for multiobjective optimization: Formulation, discussion and generalization
An overview of evolutionary algorithms in multiobjective optimization
Evolutionary Computation
Global multiobjective optimization using evolutionary algorithms
Journal on Heuristics
Intelligent Strategies for Meta Multiple Criteria Decision Making
Global multiobjective optimization with evolutionary algorithms: Selection mechanisms and mutation control
Several concepts for ε-efficiency
OR Spektrum
Adaptation in Natural and Artificial Systems
Multicriterion decision making
Cited by (30)
A comprehensive review of deterministic models and applications for mean-variance portfolio optimization
2019, Expert Systems with ApplicationsHybrid bi-objective portfolio optimization with pre-selection strategy
2017, Information SciencesRuntime analysis of a multi-objective evolutionary algorithm for obtaining finite approximations of Pareto fronts
2014, Information SciencesCitation Excerpt :These algorithms can explore the feasible spaces of multi-objective optimization problems (MOPs) to obtain uniformly distributed Pareto vectors, which has been shown by abundant numerical results [41,24,42,10,43,21,11,2,38,44,7,13,23,29,34,35]. Meanwhile, theoretical studies of convergence [26,25,16,40,32,8,1] and runtime analyses [14,26,28,31,5,6,18,19,3,15,20,4,12,33] have also been performed to explain how MOEAs function on different MOPs. By adding objectives to a well-known plateau function, Brockhoff et al. [6] found that changes in running time are caused by changes in the dominance structure.
Multiobjective Evolutionary Algorithms for Portfolio Management: A comprehensive literature review
2012, Expert Systems with ApplicationsInteractive Multiple Criteria Decision Making based on preference driven Evolutionary Multiobjective Optimization with controllable accuracy
2012, European Journal of Operational ResearchCitation Excerpt :A way of employing the above bounds (outcome assessments) in MCDM is proposed in the next section and illustrated by a numerical example in Section 7. To derive SL one can use any EMO algorithm (cf. Michalewicz, 1996; Deb, 2001; Coello Coello et al., 2002; Hanne, 2007) satisfying or modified appropriately to satisfy condition (2). It is clear that with such a replacement formula (13) for lower bounds and formula (14) for upper bounds remain valid only if upper approximation AU does not contain elements which are dominated by some elements of N. However, this may be the case, as shown in Fig. 5, and certainly such elements are undesirable in any construct interpreted as an upper approximation of N. Condition (15) is meant to limit the domain for such elements.
The mean-variance cardinality constrained portfolio optimization problem: An experimental evaluation of five multiobjective evolutionary algorithms
2011, Expert Systems with ApplicationsCitation Excerpt :In this study we compare the effectiveness of the current state of MOEAs together with a simple steady state evolutionary algorithm on the MVCCPO problem formulated in Section 2. The following five MOEAs are tested: Strength Pareto evolutionary algorithm 2 (SPEA2) (Zitzler, Laumanns, & Thiele, 2001), non-dominated sorting genetic algorithm II (NSGA-II) (Deb, Ptratap, Agarwal, & Meyarivan, 2002), Pareto envelope-based evolutionary algorithm (PESA) (Corne, Knowles, & Oates, 2000), Niched Pareto genetic algorithm 2 (NPGA2) (Erikson, Mayer, & Horn, 2001), e-multiobjective evolutionary algorithm (e-MOEA) (Hanne, 2007). Except their popularity, these algorithms were chosen in such a way as to represent the main algorithmic aspects proposed up to know in the MOEA literature and are these aspects that are actually compared.