摘要
敞开式断路器的典型故障是由内部填充的绝缘硅脂性能下降而导致的真空灭弧室闪络,该故障早期会引发沿瓷套表面的沿面局部放电,对该局部放电的检测易受到外界噪音信号干扰,降低故障诊断精度。针对敞开式断路器这种放电类型明确而干扰信号不明的特点,提出基于非监督学习的敞开式断路器异常局部放电检测方法,解决传统监督学习样本不均衡且难获取的问题。该方法以4种局部放电图谱为对象,使用基于卷积神经网络的自编码器对输入图谱进行重构分析,将重构输出和原始输入之间的差异最小化作为学习目标,使模型学习局部放电特征。学习完成后,当新的放电图谱重构误差大于设定的阈值时,放电被认定为异常干扰。为进一步提升检测效果,采用D-S证据理论对4种局部放电图谱的结果进行融合。算例结果表明,该方法可以有效提升对断路器内部放电和干扰信号的区分准确率。所提方法可用于实时高效辨识敞开式断路器瓷套内部局部放电,剔除电晕干扰,为该设备的带电检测提供技术支撑。
The typical fault of the open circuit breaker is flashover in the vacuum arc-extinguishing chamber caused by the degradation of internally filled insulating silicone grease,which will cause partial discharge along the surface of the porcelain bushing in the early stage.The detection of this partial discharge is susceptible to external noise interference,reducing the accuracy of fault diagnosis.In view of the characteristics of the open circuit breaker with clear discharge type and unclear interference signal,this paper proposes an abnormal partial discharge detection method for the open circuit breaker based on unsupervised learning to solve the problem of unbalanced samples and scarce samples of traditional supervised learning.Taking four kinds of partial discharge patterns as the objects,this method uses an autoencoder based on a convolutional neural network(CNN)to reconstruct the input patterns,and learns the partial discharge characteristics by minimizing the difference between the reconstructed output and the original input.After finishing learning,when the new discharge pattern with reconstruction error is greater than the setting threshold,it is affirmed to be abnormal external interference.To further improve the detection effect,the paper employs D-S evidence theory to integrate the four partial discharge patterns results.The case studies show that this proposed method can effectively identify the internal partial discharge and interference signals with high accuracy for the circuit breaker,which can be used to identify the internal partial discharge in the porcelain cylinder in real time and provide technical support for the live detection of the devices.
作者
王录亮
李炳康
吴育毅
徐钟祝
李岩
WANG Luliang;LI Bingkang;WU Yuyi;XU Zhongzhu;LI Yan(Electric Power Research Institute of Hainan Power Grid Co.,Ltd.,Haikou,Hainan 571900,China;Key Laboratory of Physical and Chemical Analysis for Electric Power of Hainan Province,Haikou,Hainan 570311,China;Department of Electrical Engineering,North China Electric Power University,Baoding,Hebei 071003,China)
出处
《广东电力》
2023年第7期77-83,共7页
Guangdong Electric Power
基金
海南电网有限责任公司科技项目(073000KK52200011)。
关键词
断路器
局部放电
异常检测
重构误差
非监督学习
circuit breaker
partial discharge
anomaly detection
reconstruction error
unsupervised learning
