AN IMPROVEMENT OF CLUSTERING-BASED OBJECTIVE REDUCTION METHOD FOR MANY-OBJECTIVE OPTIMIZATION PROBLEMS

Abstract

Multi-Objective Evolutionary Algorithms (MOEAs) have been widely used for solving multi-objective optimization problems (MOPs). As the number of objectives is more than three, MOPs are regarded as many-objective optimization problems (MaOPs). They bring difficulties to existing MOEAs in terms of deterioration of the search ability, requirement of exponential population size, visualization of results, and especially computation cost. Although in real world, numerous optimization problems are considered as MaOPs but they contain redundant objectives. With these problems, removing the redundant objectives can alleviate these difficulties of MaOPs. Recently, clustering has been applied to remove redundant objectives such as Mutual Information and Clustering Algorithm (MICA-NORMOEA) and Objective Clustering-Based Objective Reduction Algorithm (OC-ORA). However, these clustering-based algorithms are computationally complex and they also manually tune a set of parameters. This paper proposes a clustering-based objective reduction method (COR) for MaOPs. The proposed method is more efficient than existing ones and it can automatically determine these parameters. Moreover, the proposed method still achieves comparable results with existing methods.