遺伝的アルゴリズムの新しい局所改善法の提案

抄録

The genetic algorithm (GA) is one of the basic models of evolution and is one of the effective tools for constructing evolvable/adaptive complex systems. One of the difficulties of the GA is its low performance in improving local portions of chromosomes. The authors have proposed a local improvement mechanism for the GA which is efficient for the local improvement. This paper proposes a new algorithm which introduces a hill-climbing to the GA with the local improvement mechanism. The proposed method is very efficient in improving the local portions of chromosomes by doing hill-climbing of each portion of chromosomes. To demonstrate the effectiveness of the proposed method, evolutions of arm morphology and throwing motion for throwing a ball under limited link torques is simulated.