( 天津科技大学 计算机科学与信息工程学院, 天津 300222)
摘 要:在现有的基于空间约束的空间聚类算法DBCluC和DBRS+等的研究和比较基础上,提出了一种新的处理物理约束的基于密度的空间聚类算法——DBCluC+。该算法在DBCluC算法基础上,采用网络拓扑结构建模通达对象,并增加通达对象访问点的宽度属性,从而采用约束距离(constrained distance)代替简单的欧几里德距离或障碍距离(obstacle distance)作为相异度的度量标准。理论分析和实验结果表明,DBCluC+算法不仅具有密度聚类算法的优点,而且聚类结果比传统的处理通达约束的聚类算法更合理,也更加符合实际情况的需要。
关键词:聚类; 约束距离; 网络拓扑; 障碍距离; 无向图
Clustering spatial data in presence of physical constraints:
(College of Computer Science & Information Engineering, Tianjin University of Science & Technology, Tianjin 300222, China)
Abstract:Based on learning the current spatial clustering algorithms in the presence of constrains, such as DBCluC, DBRS+ and so on, this paper proposed a new method of densitybased spatial clustering called DBCluC+ which could handle the spatial constrains in a new way. In DBCluC+, the algorithm used network topology to model the facilitator and added a attribute named width of access point, so it used the constraint distance to replace the Euclidean distance or obstacle distance in DBCluC to as the criterion of the dissimilarly. Both theory analysis and experimental results confirm that the experiments show that the new proposed approach not only has the advantages of densitybased clustering algorithms, but also takes advantage of the constraint distance to make the results more reasonable than traditional ways, and the model of algorithm is according to the need of application. ......