基于油中溶解气监测的变压器在线半监督故障诊断方法研究  

Online Semi-supervised Fault Diagnosis of Transformer Based on Dissolved Gas in Oil

作　　者：朱庆东 朱文兵 王浩哲 顾朝亮 徐冉 朱孟兆 ZHU Qingdong;ZHU Wenbing;WANG Haozhe;GU Zhaoliang;XU Ran;ZHU Mengzhao(State Grid Shandong Electric Power Company Electric Power Research Institute,Jinan 250002,Shandong Province,China)

机构地区：[1]国网山东省电力公司电力科学研究院,山东省济南市市250002

出　　处：《电网技术》2023年第3期1031-1037,共7页Power System Technology

基　　金：国网山东省电力公司科技项目(变压器综合监测智能预警和远程诊断关键技术研究及应用,52062620004L)。

摘　　要：变压器是电力系统中的核心设备,其运行的可靠性直接影响着整个电力系统的稳定与安全,因此对变压器运行状态进行实时分析并进行准确的故障诊断非常重要。针对变压器运行状态数据难以收集、故障数据缺乏而导致故障分析模型泛化能力差的问题,该文提出了一种半监督流形嵌入(semi-supervised manifold embedding,SSME)学习的变压器在线故障诊断方法。该方法使用变压器油中H2、CH4、C2H6、C2H4和C2H25种不同气体的浓度特征和运行状态类别的有限样本,联合大量在线监测获得的气体浓度样本数据,建立一种在线的半监督故障诊断模型来分析变压器的运行状态,该模型能在0.1s内完成700条在线监测数据的状态检测,其性能可以达到在线诊断的要求。结合实例,对所设计的变压器故障诊断模型的准确性和诊断效率进行了对比分析实验。实验结果表明,该文提出的方法的故障诊断准确率高于经典的深度置信网络(deep belief network,DBN)、k-近邻(k-nearby network,KNN)和随机森林(random forest,RF)分类算法。该方法能为电力变压器的稳定运行提供了有效参考依据。Considering the power transformer is one of the key components of the power system, its reliability directly affects the stability and safety of the entire system. Therefore, it is important to analyze the transformer’s operating status in real-time and perform an accurate fault diagnosis. A semi-supervised manifold embedding(SSME) method for online transformer fault diagnosis was presented in this paper in response to the problem that it is difficult to collect transformer operating status data, and the unreliability of fault analysis models due to the lack of fault data. This model is developed based on the concentration characteristics of five different gases in the transformer oil, including H2, CH4, C2H6,C2H4and C2H2. The gas concentration sample data were obtained from online monitoring in order to analyze the operation status of the transformer. In this model, the status detection of 700 online monitoring data can be performed within 0.1 seconds, thereby meeting the requirements of online analysis. Experimental comparisons of different transformer fault diagnosis models are performed to determine its prediction accuracy and efficiency. According to our experimental results, the proposed method has higher fault diagnosis accuracy than the classical deep belief network(DBN), K-Nearest Neighbor(KNN), and Random Forest(RF)methods. This method can serve as an effective reference for the safe operation of power transformers.

关 键 词：在线监测 变压器 故障诊断 半监督学习 

分 类 号：TM721[电气工程—电力系统及自动化]