融合相似度计算与改进遗传算法的聚类分析  

An Optimization Clustering Algorithm Based on Multi-Population Genetic Simulated Annealing Algorithm

作　　者：冯劲 姚远[1] FENG Jin;YAO Yuan(Beijing Institute of Technology,Zhuhai,Zhuhai Guangdong 519088,China)

机构地区：[1]北京理工大学珠海学院,广东珠海519088

出　　处：《计算机仿真》2020年第9期226-230,共5页Computer Simulation

基　　金：广东省科技计划项目(2013B051000044);广东高校省级重点平台和重大科研项目(2017KQNCX247);广东省青年创新人才项目(2016KQNCX203);2018国家级大学生创新训练计划资助项目(201813675001)。

摘　　要：针对模糊C-均值聚类分析(FCM)易陷入局部最小值以及对初始聚类中心敏感度过大的缺点,首先使用一种基于密度的DBSCAN算法,通过计算数据间距离与密度的方法确定聚类数,同时在遗传模拟退火算法(SAGA)的基础上,提出了基于多种群遗传模拟退火算法的聚类分析。首先对FCM进行分析与评价,提出FCM在确定聚类数与聚类过程方面的不足;然后针对FCM中的不足选择ST-DBSCAN算法确定聚类数,同时对遗传模拟退火算法进行研究,加入多种群并行遗传思想对遗传模拟退火算法进行优化;最后将FCM与多种群遗传模拟退火算法有机结合,优化聚类过程。实验结果表明,上述算法有较好的全局搜索能力与收敛能力,同时在聚类效果与稳定性上较传统聚类算法有一定的优势。As fuzzy C-means clustering algorithm(FCM)is easily trapped in local minima and is very sensitive to initialization of center of clusters,this paper first used a density-based DBSCAN algorithm(ST-DBSCAN)to de⁃termine the number of clusters by calculating the distance and density between data.Simultaneously,based on genetic simulated annealing algorithm(SAGA),this paper proposed an optimization clustering algorithm based on multipopulation genetic simulated annealing algorithm.Firstly,we analyzed and evaluated the FCM,pointing out the short⁃comings of FCM in determining the number of clusters and clustering process.Aiming at these disadvantages,we se⁃lected ST-DBSCAN algorithm to determine the number of clusters,studied the genetic simulated annealing algorithm,and optimized the algorithm by adding the idea of multi-population parallel genetic concept.Finally,FCM was com⁃bined with multi-population genetic simulated annealing algorithm to optimize the clustering process.The experimen⁃tal result shows that this algorithm shows better global search capability and convergence ability.Meanwhile,in terms of clustering results and stability,it has an advantage over traditional clustering algorithm.

关 键 词：聚类数 聚类过程 多种群遗传模拟退火算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]