基于SCADA系统的风电机组KNN故障状态监测研究  

KNN fault state monitoring of wind turbine based on SCADA system

作　　者：胡龙舟 李韬睿 吴頔 丰金浩 覃思航 Hu Longzhou;Li Taorui;Wu Di;Feng Jinhao;Qin Sihang(Hubei UHV Company Communication Center of State Grid,Hubei Wuhan,430050,China;State Grid Wuhan Power Supply Company,Hubei Wuhan,430050,China)

机构地区：[1]国网湖北超高压公司信通中心,湖北武汉430050 [2]国网武汉供电公司,湖北武汉430050

出　　处：《机械设计与制造工程》2025年第1期91-94,共4页Machine Design and Manufacturing Engineering

摘　　要：为了进一步提高风电机组设备的故障状态监测精度,采用2 MW风电机组数据采集与监视控制(SCADA)系统进行数据采集测试,通过K最近邻(KNN)算法综合评价了风电机组故障状态下的全工况参数。以统计过程控制(SPC)与滑动窗口结合的方法获得异常率结果,实时监测风电机组齿轮箱的实际运行状态。研究结果表明:采用优化距离度量方式能够实现预测精度的大幅提高;离群点剪辑使训练集失去一定比例的有效训练样本,但提升了运算效率;设定合适的相似剪辑阈值可以使预测精度提高0.62%,经过两次剪辑处理后相对剪辑前精度降低2.48%、运算效率提高20.92%。In order to further improve the monitoring accuracy of the fault state of the wind turbine equipment,the data acquisition and monitoring control(SCADA)system of the 2 MW wind turbine is used for data acquisition and testing.The K-nearest neighbor(KNN)algorithm is used to comprehensively evaluate the parameters of all working conditions under the fault state of the wind turbine.The abnormal rate is obtained by the method of statistical process control SPC and sliding window,and the actual running state of the gear box is monitored in real time.The results show that the optimized distance measurement can greatly improve the prediction accuracy.The outlier clipping training set also loses a certain proportion of effective training samples,which improves the operation efficiency.Setting the similar clip threshold can improve the prediction accuracy by 0.62%,reduce the accuracy by 2.48%and increase the operation efficiency by 20.92%after two clips.

关 键 词：风电机组 齿轮箱 状态监测 最近邻算法 数据采集与监视控制系统 

分 类 号：TM315[电气工程—电机]