摘要
针对水电机组运行数据异常诱因定位效率低、依赖人工经验的问题,提出一种基于单类支持向量机(OCSVM)与矩阵分解(MF)的融合算法。该方法首先利用OC-SVM对机组时序运行数据进行无监督异常检测;在识别出异常时段后,再应用MF算法挖掘多源监测数据中隐含的关联关系,快速推荐最相关的异常诱因测点。以某抽蓄机组下导摆度异常实例验证,该方法能准确捕捉异常事件,并从38个测点中有效推荐出8个关键关联测点,为现场排查指明方向,极大提升了故障诊断效率。结果表明,所提OC-SVM-MF算法为水电机组状态异常的高效、自动化溯源提供了新思路。
In view of the low efficiency in locating the causes of abnormal operating data of hydropower units and reliance on manual experience,a fusion algorithm based on one-class support vector machine(OC-SVM)and matrix decomposition(MF)is proposed.In this method,OC-SVM is firstly used for unsupervised abnormality detection on the time series operation data of the units;after identifying the abnormal period,the MF algorithm is applied to mine the implicit correlation in the multi-source monitoring data and quickly suggest the most relevant measuring points of the cause of abnormality.Taking the abnormal lower guide bearing swing of a pumped storage unit as an example,it is verified that this method can accurately capture abnormal events and effectively suggest 8 key relevant measuring points from 38 measuring points,providing a direction for on-site troubleshooting and greatly improving the efficiency of fault diagnosis.The results show that the proposed OC-SVM-MF algorithm provides a new method for efficient and automatic traceability of abnormalities of hydropower units.
作者
吴昊
刘轩
雷俊雄
张之皓
邓文涛
聂靓靓
WU Hao;LIU Xuan;LEI Jun-xiong;ZHANG Zhi-hao;DENG Wen-tao;NIE Liang-liang(Maintenance and Test Branch of China Southern Power Grid Peak Shaving and Frequency Modulation Power Generation Co.,Ltd.,Guangzhou 510000,China)
出处
《水电站机电技术》
2026年第2期5-9,135,共6页
Mechanical & Electrical Technique of Hydropower Station
关键词
水电机组
异常检测
诱因推荐
单类支持向量机
矩阵分解
hydroelectric generator unit
abnormality detection
suggestions on causes
one-class support vector machine
matrix decomposition
