作　　者： ; ;

机构地区： 华南理工大学计算机科学与工程学院

出　　处： 《计算机学报》 2012年第2期193-201,共9页

摘　　要： 给出一种新的优化算法:球隙迁移法.该方法不是已有方法的融合或改进,它利用搜索过程中积累的极小点分布信息形成球隙,以此启发、指导后来的搜索区域,不但逃离了当前局部极小,还能有效地避免重复历史上的多个局部极小.目前的智能算法中,勘探和开采行为相耦合,球隙法实现了勘探与开采的分离,避免了相互干扰,减小了代价,对变量耦合对象的优化效果好.文中证明了球隙法能在有限计算次数内确定地找到连续函数的全局最优. This paper presents a new optimization algorithm called sphere-gap transferring algorithm. The algorithm, which is not made by improving or combining existing methods, uses the collected information of the distribution of the minima found by searching to construct spheregaps to illumine and guide later searching range. The sphere-gap transferring algorithm can not only escape the current local minimum but also avoid the found local minima. In current intelligent algorithms, exploration couples exploitation. The sphere-gap transferring algorithm separates exploration from exploitation to avoid interference and to reduce cost. The algorithm is good at coupling objects. This paper proves that the sphere-gap transferring algorithm can determinately find the globally optimal solution of a continuous function in the limited number of computation.

关 键 词： 优化 全局极小 局部极小 连续函数

领　　域： [自动化与计算机技术] [自动化与计算机技术]