Dynamic adaptation of decomposition vector set size for MOEA/D

Abstract

The Multi Objective Evolutionary Algorithm based on Decomposition (MOEA/D) is a popular algorithm for solving Multi Objective Problems (MOP). The main characteristic of MOEA/D is to use a set of weight vectors to break the MOP into a set of single-objective sub problems. It is well known that the performance of MOEA/D varies greatly depending on the number of weight vectors. However, the appropriate value for this hyper-parameter is likely to vary depending on the problem, as well as the stage of the search. In this study, we propose a robust MOEA/D variant that evaluates the progress of the search, and deletes or creates weight vectors as necessary to improve the optimization or to avoid search stagnation. The performance of the proposed algorithm is evaluated on the DTLZ and ZDT benchmark. We observed that the proposed method without needing to explicitly choose the number of weight vectors is equivalent to MOEA/D with fine tuned vectors and superior than MOEA/D with poorly tuned vectors.