基于并行模糊C-均值聚类的风电机组发电机故障诊断研究  

FAULT DIAGNOSIS OF WIND TURBINE ALTERNATOR BASED ON PARALLEL FUZZY C-MEANS CLUSTERING ALGORITHM

作　　者：孙鹤旭[1] 孙泽贤[1] 张靖轩 Sun Hexu;Sun Zexian;Zhang Jingxuan(Institute of Control Science and Engineering,Hebei University of Technology,Tianjin 300130,China;College of Electrical Engineering,North China University of Science and Technology,Tangshan 063000,China)

机构地区：[1]河北工业大学控制科学与工程学院,天津300130 [2]华北理工大学电气工程学院,唐山063000

出　　处：《太阳能学报》2020年第3期8-14,共7页Acta Energiae Solaris Sinica

基　　金：河北省科技计划(17214304D);天津市科技支撑项目(14ZCDZGX00818)。

摘　　要：提出一种基于并行化的改进模糊C-均值聚类的风电机组发电机故障诊断方法。首先通过邻近聚类算法确定数据集可能的最大类簇数Cmax,以Cmax为上限,利用改进的模糊C-均值算法,以BWI(between-within index)指标为聚类有效性判别指标,提出新的确定最佳聚类数的方法。并结合Spark内存处理技术,将其应用在风电机组发电机的故障诊断中,通过UCI机器学习数据库数据集以及风电监测实时数据的实验测试,表明该算法不仅能准确判别发电机的故障模式,并且能更好地处理电力系统的海量数据。A new method based on improved FCM(fuzzy C-means)on fault diagnosis for wind turbine is proposed. First,the affinity propagation clustering algorithm is used to determine the max number of clusters. Then the improved FCM and the criterion of BWI(Between-Within Index)are combined to determine the optimal number of clusters of a dataset. The parallel FCM based on Spark is used to detect the fault in the wind turbine. The improved FCM algorithm are evaluated by studying the UCI datasets and camparing with the real-time data from the experiments,the results show that the method can effectively improve the accuracy fault diagnosis and can better process massive data in power system.

关 键 词：故障诊断 聚类 风电机组 发电机 风电监测 SPARK 

分 类 号：TM914[电气工程—电力电子与电力传动]