基于K均值聚类与局部离群因子算法的故障检测研究

Fault Detection Based on K-means Clustering and Local Outlier Factor Algorithm

针对化工生产过程的多工况、数据多模态问题,提出一种基于K均值聚类的局部离群因子故障检测方法。首先利用K均值聚类算法对多模态工业数据进行聚类,将各个模态的数据分离出来,然后运用局部离群因子算法在各个模态下单独建立模型,并且确定各个模态下的局部离群因子控制限。检测时首先判断样本属于哪一类,然后在相应类别下求取局部离群因子值并与此类别下的控制限进行比较,确定是否为故障数据。将此方法运用到TE过程的多模态数据中,并且将此方法与单独应用局部离群因子算法做故障检测对比,结果表明:所提算法可以大幅提高故障的检测率。

Considering the multi-condition and the data’s multi-model of chemical production process, a K-means clustering-based local outlier factor(LOF) fault detection method was proposed. Firstly, having the K-means clustering algorithm adopted to cluster multimodal industrial data and to isolate each modal data, then having LOF algorithm employed to establish the model under different modal model separately and to determine each modal’s LOF control limit. In the detection, having the sample judged to determine its kind and then to obtain its local outlier factor value under the corresponding categories and to compare it with the control limit under this category so as to make sure whether it is a fault data. In this paper, applying this method to the TE process of multimodal data and comparing its result with that of the fault detection through using local outlier factor alone show that, the K-means clustering algorithm can greatly improve the fault detection rate.