Parallelization of the Conical Area Evolutionary Algorithm on Message-Passing Clusters

The Conical Area Evolutionary Algorithm (CAEA) has exhibited significant performance for bi-objective optimization problems. In this paper, a parallel partially evolved CAEA (peCAEA) on message-passing clusters is proposed to further reduce the runtime of the sequential CAEA for bi-objective optimization. Each island maintains an entire population but is responsible for evolution of only a portion of the population. Further, the elitist migration adopted in order to share information among islands and speed up the evolutionary process. Additionally, a dynamic directional migration topology presented here to obtain a satisfactory balance between convergence speed and communication costs. Experimental results on ZDT test problems indicate that the peCAEA obtains satisfactory solution quality with good speedup on clusters.