Abstract: As for the problem of a large quantity of alarm data disposing and analysis of urban rail transit equipment, the approach of data mining is applied and clustering feature tree is designed to establish multi-level clustering analysis model based on fault collect, combining with characters of equipment fault. The high frequency fault clusters are attained, aiming to identify association rules of factors that alarms arouse. The fault pre-alarm model is proposed through constructing the hierarchical Bayesian Network（BN） for equipment alarm, to calculate and infer equipment failure through value probability. The method of distinguishing associated pre-alarm and its implement strategy are also discussed. A period of monitoring data of certain urban rail transit equipment serves as an example, according to results, the calculated fault coordination between equipment and related systems accords with actual ones. The result indicates the method is helpful to speculate operation equipment alarms and quickly locate in potential risks, and it can make effective decision for safety management.

Key words: urban rail transit, equipment fault, pre-alarm, clustering, Bayesian Networks（BN）

摘要： 针对城市轨道交通的海量监控报警数据分析难度大的问题，结合设备故障特征，应用数据挖掘方法定义聚类特征树，建立基于故障集的层次聚类分析模型，取得故障要素的高频聚类并识别关联规则。通过构建故障贝叶斯层次网络，提出故障预警模型，计算取值概率推断预警值，进一步探讨了关联性预警的识别和应用策略。采用某城市轨道交通的阶段报警数据实例验证，表明识别出的故障关联规则与实际较好地符合，利用该方法推断故障预警能快速定位风险隐患，为安全管理提供有效的决策支持。

关键词: 城市轨道交通, 故障, 预警, 聚类, 贝叶斯网络