Self-adaptive Multi-population Genetic Algorithms

• Main Authors: Wen-Yean Lee, 李文淵
• Other Authors: Wen-Yang Lin
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/92563234761465440722

碩士 === 義守大學 === 資訊工程學系 === 89 === Multi-population genetic algorithm (MGA), a macro-evolutionary search paradigm derived from the punctuated equilibrium theory, has been recognized as a more effective model than traditional single-population genetic algorithms (SGAs). This model is based on a multi-population structure within which each sub-population evolves independently and is occasionally punctuated by inter-population migration. It has been shown that such population structure and isolated evolution would wide the search space and is immune from premature convergence. Despite of these advantages over SGAs, the performance of MGAs, like SGAs, is heavily affected by a judicious choice of evolutionary schemes and parameter settings, and the choice is dependent on the problem as well. This thesis is an effort to cope with the problem of automating parameter settings for MGAs. We propose a framework for studying the self-adaptation of MGAs and address various self-adaptive schemes that can be incorporated into the evolution. Though not yet comprehensive, the result of our work has illustrated the effectiveness of self-adaptation for MGAs and paved the way for this unexplored area.