Enhanced micro genetic algorithm-based models for multi-objective optimization
Loading...
Date
2014-09
Authors
Tan, Choo Jun
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Multi-objective Optimization Problems (MOPs) entail multiple conflicting objectives to be
satisfied simultaneously. As such, a set of alternative solutions that is able to satisfy all objectives
with respect to the Pareto optimality principle is desired. Besides that, the quality of good
MOP solutions needs to strike a balance between convergence and diversity against the true
Pareto front (i.e. distribution of the ideal Pareto optimal solutions). This research is concerned
with how evolutionary algorithms can be employed to undertake MOPs with good convergence
and diversity properties of the solutions with respect to the true Pareto front. Specifically, three
evolutionary models based on the micro Genetic Algorithm (mGA) have been developed in
a sequential manner for undertaking MOPs. Firstly, a Modified mGA (MmGA) model is introduced.
MmGA aims to search for the Pareto optimal solutions efficiently and improve the
convergence score of the solutions towards the Pareto front in tackling MOPs. Secondly, an ensemble
of MmGA models is proposed to improve the robustness of individual MmGA models
and, at the same time, to accelerate the convergence score of the solutions towards the Pareto
front. Thirdly, to take both convergence and diversity scores of the Pareto optimal solutions
into consideration, a multi-agent system that utilizes the Trust-Negotiation-Communication
(TNC) reasoning scheme is exploited. Multiple performance indicators are incorporated into
the TNC-based MmGA model, in order to achieve good convergence and diversity scores. The
computational time complexity of three proposed evolutionary models is examined using the
O-notation analysis. It is found that, while the computational time complexity of the three proposed
models is higher than that of mGA, all proposed models have the same computational
xxiii
time complexity with that of the Non-dominated Sorting Genetic Algorithm II (NSGA-II),
which is embedded into all proposed models to improve their performances. Based on a number
of MOP benchmark problems, all proposed models are able to yield favorable results as
compared with those from their predecessors in term of convergence and diversity scores of
the solutions. A number of experiments with multi-objective classification problems also indicate
improvement of the classifier performances and, at the same time, reduction of the number
of features used in classification, as compared with the results from standard classifiers as well
as from other methods published in the literature. The potential of the proposed MmGA-based
models in undertaking practical problems has also been demonstrated using four real-world
MOPs. All performance indicators of the proposed MmGA-based models are quantified using
the bootstrap statistical method. The outcomes positively ascertain the usefulness of the
proposed evolutionary models in undertaking MOPs.
Description
Keywords
Algorithm-based Models