Evolutionary algorithms and multi-objectivization for the travelling salesman problem

Abstract

This paper studies the multi-objectivization of single-objective optimization problems (SOOP) using evolutionary multi-objective algorithms (EMOAs). In contrast to the single-objective case, diversity can be introduced by the multi-objective view of the algorithm and the dynamic use of objectives. Using the travelling salesman problem as an example we illustrate that two basic approaches, a) the addition of new objectives to the existing problem and b) the decomposition of the primary objective into sub-objectives, can improve performance compared to a single-objective genetic algorithm when objectives are used dynamically. Based on decomposition we propose the concept "Multi-Objectivization via Segmentation" (MOS), at which the original problem is reassembled. Experiments reveal that this new strategy clearly outperforms both the traditional genetic algorithm (GA) and the algorithms based on existing multiobjective approaches even without changing objectives.