On Performance Improvement Based on Restart Meta-Heuristic Implementation for Solving Multi-objective Optimization Problems

Abstract

One of the possible goals of multi-objective optimization is finding a set of non-dominated solutions or, in other words, a Pareto set approximation. Population-based algorithms, in particular, genetic algorithms, are widely used for this purpose because they deal with a set of alternative solutions, which might be helpful when a number of trade-off points should be obtained. To get a representative approximation, various regions of a search space should be explored. However, during the algorithm execution a search might be stuck in some areas. Therefore, in this article we present a new restart operator for multi-objective genetic algorithms which can prevent a search from stagnating, help to explore new regions and, as a result, improve the algorithm performance significantly. In our proposal we answer the two crucial questions of a restarting concept which are when to restart an algorithm and how to use previously found solutions. We introduce the algorithm independent restart operator, even though in this work we investigate it in combination with a certain MOGA. The experimental results prove the high effectiveness of the modified MOGA with the incorporated restart operator in comparison with the conventional one.