群搜索优化算法中角色分配策略的研究  

Study on Role Assignment Strategies of Group Search Optimizer

作　　者：汪慎文[1,2] 丁立新[1] 谢承旺[3] 谢大同[1] 舒万能[1] 杨华[4] 

机构地区：[1]武汉大学软件工程国家重点实验室,武汉大学计算机学院,武汉430072 [2]石家庄经济学院信息工程学院,石家庄050031 [3]华东交通大学软件学院,南昌330013 [4]贵州师范大学数学与计算机学院,贵阳550001

出　　处：《小型微型计算机系统》2012年第9期1938-1943,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60975050,61070243,61165004)资助;高等学校博士学科点专项科研基金项目(20070486081)资助;中央高校基本科研业务费专项基金项目(6081014)资助;河北省科技支撑计划项目(11213587)资助;江西省自然科学基金项目(20114BAB201025)资助;江西省教育厅科技项目(GJJ12307)资助

摘　　要：群搜索优化算法把群体中的个体分为三种角色:发现者,加入者和游荡者.算法选择最优个体作为唯一的发现者,采用随机决策策略分配加入者和游荡者.该策略过于盲目,算法性能也较差.针对这一缺点,本文考虑个体与发现者的分布关系,以发现者为球心,聚集在球心(或圆心)周围的个体均为加入者,散布于球外的个体为游荡者.而球的半径大小的确定,提出两种不同的方法:自适应半径策略和固定加入者个体数量策略.前者取个体到发现者的平均距离的K倍作为球的半径,后者规定离发现者最近的一定数量的个体被认为分布于球内的加入者.通过实验表明,这两种策略提高了种群多样性的同时,还保证了算法快速收敛到最优解.Group Search Optimizer （GSO） assumes that group members include producers, scroungers and rangers. The algorithm of- ten chooses the best member as the producer. However, the Group Search Optimizer applies random strategy for the role assignment of scrounger and ranger, which is a blind search and leads to bad results. So, based on the spatial distribution between the members and the producer, we assume that all individuals inside the sphere with producer as the center are scroungers and rangers are ones out- side the sphere. In order to determine the radiu of the sphere, we propose two new methods： strategy of self-adaptation radius and strategy of fixed scrounger number. The former method takes k times of the average distance from the members to the producer as the radius of the sphere. The latter one assumes that certain indviduals with the nearest distance from producer are scourngers. Results shows that the novel algorithms achieve faster convergence and improve the diversity of individuals.

关 键 词：群体智能 群搜索优化算法 发现 加入模型 角色分配 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]