摘要
为了提升差分演化算法对局部最优的逃逸能力和避免早熟收敛,设计了一种邻域结构为复杂网络的差分演化算法(CNS-DE)。该算法将复杂网络上每一个节点定义为一个计算个体,节点间的连接关系决定了个体间的交互结构。CNS-DE的差分策略主要基于节点(个体)的邻居关系定义,该策略有利于保持群体的多样性,充分利用了群体分布特性。在函数寻优的经典数据集上,将CNS-DE与传统差分算法进行了对比。结果表明,该算法能有效避免陷入局部最优,有效改善了早熟现象,对解的质量有较大幅度提高。
In order to improve the capability of escaping local optimum for the differential evolution algorithm, and avoid pre- mature convergence, this paper designed a new algorithm named CNS-DE. The algorithm adopted a complex network as its spatial structure. Specifically, CNS-DE put an individual on one node of the network; the individual evolved by mainly inter- acting with its neighbors on the network. Based on nodes＇ （ individuals＇ ） neighbor relationship, this paper proposed a new differential strategy for CNS-DE. The policy fully used the distribution of group and is conductive to maintaining population di- versity. On the classic dataset for the tasks of function optimization, a series of experimental results of CNS-DE and DE show that the new algorithm can effectively avoid getting into local optimum, and effectively improve the precocious phenomenon. In addition, it greatly increases the quality of solutions.
出处
《计算机应用研究》
CSCD
北大核心
2016年第2期370-374,共5页
Application Research of Computers
基金
国家自然科学资金资助项目(71271067)
关键词
复杂网络
演化计算
差分算法
函数优化
空间结构
complex network
evolution computation
differential algorithm
function optimization
spatial structure
