基于递归聚类算法和漂移检测的发电厂设备故障诊断  

Fault diagnosis of power plant equipment based on recursive clustering algorithm and drift detection

作　　者：农振昌 王伟 吴政 刘丰 贾高祥 NONG Zhenchang;WANG Wei;WU Zheng;LIU Feng;JIA Gaoxiang(Longtan Hydroelectric Power Plant,Longtan Hydropower Development Co.,Ltd.,Nanning 530000,China)

机构地区：[1]龙滩水电开发有限公司龙滩水力发电厂,广西南宁530000

出　　处：《电子设计工程》2024年第17期36-39,44,共5页Electronic Design Engineering

基　　金：2022年龙滩水电开发有限公司龙滩水力发电厂信息化项目(CDT-LTHPC-X-2374)。

摘　　要：为了解决复杂非平稳动态系统中的故障问题,需要一种自适应故障诊断方法来应对被监测过程的变化,基于此,提出了一种将递归聚类算法与漂移检测方法相结合的进化模糊分类器,并应用于故障诊断。在该方法中,递归聚类算法的更新不仅依赖于相似性度量,还依赖于输入数据流中的监测变化。使用漂移检测方法在数据发生变化时检测故障发生时间。在模糊规则中采用多元高斯隶属函数,避免了变量间相互作用时的信息丢失。对提出方法在直流传动系统故障诊断中的应用进行了验证,在模拟数据中加入异常值和噪声来评估故障诊断模型的鲁棒性。结果表明,该分类器在存在异常值和噪声的情况下,仍能以较高的性能对所有故障进行检测和分类,具有较高的故障隔离率。In order to solve the faults in complex non⁃stationary dynamic systems,an adaptive fault diagnosis method is needed to cope with the changes of the monitored process.Based on this,an evolutionary fuzzy classifier combining recursive clustering algorithm and drift detection method is proposed in this paper.In this method,the update of the recursive clustering algorithm depends not only on the similarity measure,but also on the monitoring changes in the input data stream.The drift detection method is used to detect the time of failure when the data changes.Multivariate Gaussian membership function is used in fuzzy rules to avoid information loss when variables interact with each other.The application of the proposed method in DC transmission system fault diagnosis is verified,and the robustness of the fault diagnosis model is evaluated by adding outliers and noise to the simulation data.The results show that the classifier can detect and classify all faults with high performance and high fault isolation rate in the presence of outliers and noise.

关 键 词：聚类算法 漂移检测 鲁棒性 故障诊断 

分 类 号：TN713[电子电信—电路与系统] TM73[电气工程—电力系统及自动化]