進化型多目的最適化手法の基礎的検討と並列化

概要

論文の詳細を見る
The aim of our study is to clarify basic performance evolutional methods for multi-objective optimization, and to parallel these algorithms to shorten the calculation time using the parallel processing. We select three evolutional methods, i.e., MOGA, NSGA-II and SPEA-II, for above aim. First, we analyze experimental results by Deb's four multi-objective problems. Then, we clarified that SPEA-II and NSGA-II are with the same search efficiency. On the other hand, as for CPU time, MOGA is the fastest, and SPEA-II is inferior for MOGA a little, and NSGA-II takes about twice of MOGA. After considering accuracy, a property of frontier and the CPU time of the Pareto solution overall, it turned out that SPEA-II is the best. Finally, with our PC cluster, applying a master slave model to above three methods, we confirm the effectiveness of PC cluster and SPEA-II has the best performance. Also, we clarify importance of adoption of appropriate parallel algorithm depending on a problem.
2007-03-20