一种具有领导机制的混合蛙跳优化算法  

Shuffled frog leaping algorithm with leadership mechanism

作　　者：张桂珠[1] 胥枫 赵芳[1] 吴德龙[1] 

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机应用研究》2014年第7期1984-1988,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61170120);江苏省自然科学基金资助项目(BK2011147)

摘　　要：针对标准混合蛙跳算法在高维复杂函数优化时易陷入局部极值的缺陷进行了研究,提出了一种具有领导机制的混合蛙跳算法。该算法借鉴"差别对待"思想将青蛙种群分类,每类青蛙采用不同的位置更新公式进行位置更新,并利用高斯变异取代随机产生新青蛙的操作,避免算法盲目搜索;模仿自然界的领导作用,引入粒子群种群在青蛙种群最优个体附近进行局部振荡搜索。为了检验新算法的寻优性能,采用六种典型的连续函数对其以及其他混合蛙跳算法进行收敛精度和收敛速度方面的对比测试。实验结果表明,新算法具有较高的收敛精度和较快的收敛速度,有效地弥补了标准混合蛙跳算法易早熟收敛的缺陷,适用于高维复杂函数的优化。Standard shuffled frog leaping algorithm（SFLA） easily trapped into local optima when it dealt with high dimensional complex functions,in order to overcome the drawbacks,this paper did some research and proposed an improved SFLA called a shuffled frog leaping algorithm with leadership mechanism（LMSFLA）. The algorithm learned from the thought of differential treatment and divided frog populations into categories,each category updated their location with different location update formula. Furthermore,the algorithm adopted Gaussian mutation to replace the original SFLA update strategy,which avoided blind search. Last,as imitating leadership role in nature,the algorithm introduced particle swarm optimization（ PSO）populations into SFLA populations to do local oscillation search nearby global optima. To show the searching performances of the algorithm,it illustrated convergence precision and convergence speed comparing tests for demonstation on the six typical continuous functions with other SFLA. The results indicate that LMSFLA has higher convergence precision and faster convergence speed,effectively makes up for the defects of standard SFLA＇s premature convergence,and is more suitable for high-dimensional optimization of complex functions.

关 键 词：混合蛙跳算法 分类 高斯变异 粒子群领导机制 全局优化 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]