关键词: 聚类分析, 离群数据挖掘, 紧密（离核）指数, 离群指数, 综合指数, 流程企业

Abstract: To monitor process industry's production, the clustering models of history conventional parameters were constructed based on feature weight's K-means algorithm. The model was used to conduct real-time inspection in parameters during equipment operation. After analyzing proportions and thresholds among different similarity factors and the whole cluster by different similarity methods, a new factor of compress/scatter was defined. Based on the conception of CBLOF(t), a new factor of outlier was brought forward to study the real-time outlier. The two factors were combined to a new factor of combination by different instances. Based on the models and the new factors, equipment operations could be effectively monitored so as to optimize equipments process and prevent defects.

Key words: clustering analysis, outlier mining, factor of compress/scatter, factor of outlier, factor of combination, process industry