基于自适应步长的萤火虫划分聚类算法  

Firefly partition clustering algorithm based on self-adaptive step

作　　者：潘晓英 陈雪静 李昂儒 赵普 

机构地区：[1]西安邮电大学计算机学院,西安710061

出　　处：《计算机应用研究》2017年第12期3576-3579,3602,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61105064;61203311);陕西省教育厅专项科研计划项目(14JK1665);厦门市科技计划项目(3502Z20141164)

摘　　要：在许多领域中,聚类是重要分析技术之一,如数据挖掘、模式识别和图像分析。针对K-means算法过度依赖初始聚类中心的选择而陷入局部最优的问题,提出了基于自适应步长的萤火虫划分聚类算法(ASFA)。利用萤火虫算法的随机性和全局搜索性找到指定数量的初始簇中心,进一步利用K-means得到精确的簇划分。在萤火虫聚类优化算法中,采用自适应步长代替原有的固定步长,从而避免算法陷入局部最优,且能获得精度更高的解。为了提高算法性能,将改进的新算法用于不同规模大小的标准数据集中,实验结果表明,ASFA与Kmeans、GAK、PSOK对比显示了更好的聚类性能和更好的稳定性及鲁棒性,与其他文献中算法相比,ASFA在寻优精度方面能取得更好的效果。In many areas,clustering is one of the most important techniques,including data mining,pattern recognition and image analysis. Due to K-means algorithm is easy to fall into the local optimum by the selection of initial clustering center,this paper proposed an improved algorithm based on the combination of firefly algorithm and K-means algorithm,which was called ASFA. By using random and global search of firefly algorithm,it initialized the original cluster centers,which could be further used to obtain more accurate clustering of K-means. In the clustering optimization algorithm,it utilized adaptive step size instead of the original fixed step size to avoid local optimization of the algorithm and obtained higher accuracy. In order to improve performance,it implemented the new algorithm in benchmark datasets of different size. The experimental results show that ASFA has better clustering performance,robustness and stability. In addition,compared with other algorithms in the literature,ASFA achieves better effect in accuracy optimization aspect.

关 键 词：萤火虫算法 K-MEANS算法 初始聚类中心 自适应步长 鲁棒性 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]