摘要
为解决传统和在线检测方法对于套管的潜伏性放电故障无法及时检测或检测灵敏度较低的问题,采用了具有较高的抗干扰能力并且可实现非接触式检测的特高频法。为分析特高频电磁波在变压器套管中的传播特性,建立了变压器套管物理模型;通过仿真计算明确了套管局部放电特高频信号的泄漏路径及实现外部无接触式检测的可行性。建立110 kV套管多类型局部放电实验平台并通过测试,得到了典型放电故障类型下对应的特高频信号波形16类时频域特征。对比分析了采用主成分分析、局部线性嵌入、核主成分分析进行特征筛选的结果,以及相关支持向量机算法进行故障诊断时的诊断结果。实验数据表明,模型在维度较高时各算法均有92%以上的准确率,这证明所选特征具有较好的代表性;在模型达到95%以上的准确率时,使用局部线性嵌入算法可将特征向量维度降低至3维、故障诊断所需数据量减小81.25%,这证明局部线性嵌入算法能够在保证高准确率的同时,高效地缩小特征维度、大幅减小数据处理量。
Aiming at the situation that the traditional and online detection methods cannot detect the latent discharge fault of bushing in time or the detection sensitivity is low,the UHF technic with high anti-interference ability and non-contact detection is adopted.To analyze the propagation characteristics of UHF electromagnetic waves in bushing,a physical model of transformer bushing is established.The leakage paths of bushing partial discharge UHF signals and the feasibility of external non-contact detection are clarified through simulation calculation.A multi-type partial discharge experimental platform for 110 kV bushing is used to test and 16 types of time-frequency domain features of UHF signal waveforms corresponding to typical discharge fault types are obtained.The results of feature screening using principal component analysis,local linear embedding and kernel principal component analysis,and the diagnosis results of fault diagnosis using related support machine algorithm are compared and analyzed.The data shows that each algorithm has an accuracy rate of more than 92%when the model dimension is high,which proves that the selected features have good representativeness.When the model reaches an accuracy rate of more than 95%,the local linear embedding algorithm can be used to reduce the features to 3 dimensions,and the amount of data required for fault diagnosis is reduced by 81.25%,which proves that the local linear embedding algorithm can effectively reduce the feature dimension and greatly reduce the amount of data processing while ensuring high accuracy.
作者
叶芃
彭璐
周迪
陈钜栋
YE Peng;PENG Lu;ZHOU Di;CHEN Judong(Ultra-high Voltage Company,State Grid Jiangxi Electric Power Company,Nanchang 330001,China)
出处
《电力科学与工程》
2022年第11期17-24,共8页
Electric Power Science and Engineering
基金
江西省电力有限公司科技项目(521823180008)。
关键词
变压器套管
局部放电
故障诊断
特高频
特征工程
模式识别
transformer bushing
partial discharge
troubleshooting
UHF
feature engineering
pattern recognition
