A Study of Distributed MOEA/D Based on Spark Framework

We carry out a careful study of Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) and propose some implementation schemas in the Spark Framework. Focusing on the setting of weight vectors, we propose two partitioning schemas, which define the distribution mode for the algorithm. The first partitioning schema is to define a partition by a group of weight vectors that are close to each other. The other schema is to distribute close weights to different partitions. Experiments in distributed framework indicate that, for most benchmarks, the schemas in distributed framework can obtain better results and better performance in expansibility.