基于相似性排挤与适应值分层计算的可持续Pareto遗传算法  

Sustainable Pareto Genetic Algorithm Based on Similarity Crowding and Hierarchical Fitness Stratification

作　　者：李少波[1] 杨观赐[1] 郭观七[2] 

机构地区：[1]贵州大学,贵阳550003 [2]湖南理工学院,岳阳414000

出　　处：《中国机械工程》2007年第14期1717-1722,共6页China Mechanical Engineering

基　　金：国家自然科学基金资助项目(50575047);教育部科学技术研究重点项目(206101);贵州省科学技术基金资助项目(黔科合J字[2005]2113号);贵州省省长基金资助项目(黔省专合字(2006)20号)

摘　　要：在研究并实现计算复杂度仅为O(Nlog2N′)的基于相似性排挤的小生境技术(NTSC)、快速适应值分层算法(FHFS)的基础上,提出了基于相似性排挤与适应值分层计算的可持续Pareto遗传算法(SPGA)。SPGA采用了进化操作种群与外部种群两个种群。外部种群用于存储当前最优解集,利用基于模糊推理机制提出的NTSC来维持种群多样度,使外部种群中存储的Pareto非劣解集均匀地逼近问题的理论最优面;采用将个体按其所处层次来精确标识个体适应能力的FHFS来辨识个体适应值,避免适应值特别高的个体抑制适应值比它低的个体。仿真优化结果表明,SPGA能够以较小的计算成本搜索到高精度的、分布均匀的Pareto非劣解集。A sustainable Pareto genetic algorithm （SPGA） was proposed for multi--objective optimization based on two techniques including the niche technique of similarity based on crowding （NT- SC） and the fast hierarchical fitness stratification （FHFS）. NTSC with computational complexity of only O（N log2 N＇） was used for maintaining the population diversity and uniform distribution of Pareto solutions while FHFS for avoiding or reducing suppression of low--fitness individuals by high--fitness individuals,genetic drift and premature convergence. Two populations were employed in the SPGA algorithm. The main population was used for genetic operation and an external population for storing current optimal solutions. Testing on benchmark multi--objective 0/1 knapsack problems demonstrates that SPGA can obtain high--quality and evenly distributed non--dominated Pareto solutions with fewer computational efforts compared to other representative algorithms.

关 键 词：遗传算法 PARETO最优性 相似性排挤 适应值分层 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP273[自动化与计算机技术—控制科学与工程]