目标综合评价关联聚类算法的改进  

Improvement of relational cluster algorithm for objective comprehensive assessment

作　　者：苏小红[1] 王亚东[1] 马培军[1] 张田文[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工业大学学报》2004年第10期1301-1303,共3页Journal of Harbin Institute of Technology

基　　金：国家自然科学基金资助项目 ( 60 2 73 0 83 )

摘　　要：利用现有的适于目标综合评价的关联聚类分析法对模糊事物进行分类评判时 ,存在阈值选取困难、聚类结果与标准对象的排列顺序直接相关、容易出现误判和漏判、先分的类聚集较多的对象而后分的类聚集较少的对象等缺点 ,影响了聚类结果的准确性和合理性 ,针对这些问题 ,提出一种改进的关联聚类算法 ,克服了原算法的不足之处 ,无论标准对象顺序如何排列、域值如何选取、先聚集哪一个类别 。When the present incidence cluster method is applied to comprehensive assessment of fuzzy objectives, some problems maybe occurred, such as the difficulty in choosing threshold, the cluster result is directly related to the sequence of arrangement of reference objectives, some fault decision and miss decision for some objectives are often made, much objectives are in former clusters but less objectives in latter clusters, and etc. These problems influence the accuracy and reasonability of cluster results. Considering these problems, an improved incidence cluster algorithm is proposed which overcomes the above defects. No matter how to choose the threshold, how to arrange the sequence of the reference objectives, which class is chosen to be clustered at first, reasonable and stable results can be obtained invariably.

关 键 词：聚类算法 对象 阈值选取 关联 域值 模糊 聚集 漏判 排列 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] TP391[自动化与计算机技术—计算机科学与技术]