Distributed Parellel MOEA/D on Spark

Abstract

The multi-objective evolutionary algorithm based on decomposition (MOEA/D) has shown remarkable performance for multi-objective optimization problems (MOPs). However, MOEA/D still consumes long time to solve MOPs with computationally intensive objective functions. This paper proposes two distributed parallel MOEA/Ds based on the popular distributed framework, Spark, to further reduce the running time of the sequential MOEA/D for MOPs. The first entirely evolved MOEA/D evolves an entire population, while the second partially evolved MOEA/D based on Spark evolves a partial subpopulation equal in size to a partition in each transformation-action process. Experimental results on DTLZ benchmark MOPs with three objectives indicate that both distributed MOEA/Ds on Spark obtains better speedup than the distributed MOEA/Ds on MapReduce and achieve the quality of solutions similar to the sequential MOEA/D.