基于差分进化算法的K-Means算法改进  

Improvement of K-Means Algorithm Based on Differential Evolution Algorithm

作　　者：刘莉莉[1] 曹宝香[1] 

机构地区：[1]曲阜师范大学信息科学与工程学院,山东日照276826

出　　处：《计算机技术与发展》2015年第10期88-92,共5页Computer Technology and Development

基　　金：山东省自然科学基金项目(ZR2009GM009);山东省科技攻关项目(2012GGB01193)

摘　　要：针对现如今传统的K-Means聚类算法所普遍存在的对初始聚类中心选择敏感且易陷入局部最优解的问题,文中将全局寻优能力较强的差分进化算法引入该算法中,其中通过采用选择结构的多模式进化方案、自适应调整的控制参数,从而提出了一种性能优良的改进的差分进化算法。同时进一步将改进的差分进化算法和K-Means聚类算法相结合,得以较好地解决了K-Means聚类算法中初始聚类中心的优化问题。通过在三种国际通用数据集上进行实验测试,最终的实验结果表明,该方法可以明显加快算法收敛速度,增强全局优化能力,并且有效提高了聚类结果的质量和稳定性。According to the defects of traditional K- Means clustering algorithm such as sensitive to the initial clustering center selection, falling into the local optimal solution easily, the differential evolution algorithm which has the rich global search ability is introduced in this paper, then an improved differential evolution algorithm with multi-model evolution scheme of selection structure and control param- eters of adaptive adjustment is presented in the meantime. The algorithm combined with K - Means algorithm has solved initial center opti- mization problem well. Experiments on the international datasets show that this method could speed up the convergence speed significant- ly ,enhancing the ability of global optimization,improving the clustering quality and stability effectively.

关 键 词：聚类算法 K—Means聚类算法 差分进化算法 进化模式 控制参数 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]