摘要
为提高边界状态变压器故障诊断的正确率,提出基于贝叶斯神经网络的变压器故障诊断方法。首先,将深度学习与贝叶斯方法结合,构建基于概率分布权重单元的贝叶斯神经网络,建立气体成分与故障状态间的概率映射关系;然后,采用变分推理方法训练贝叶斯神经网络,并基于蒙特卡罗拟合边界状态变压器的故障概率,定量评价边界状态变压器的健康情况。算例分析结果表明,所提方法对边界状态变压器的故障诊断能力更强,且对样本数据误差具有良好的鲁棒性。
To improve the accuracy of transformer fault diagnosis in a boundary state,a transformer fault diagnosis method based on Bayesian neural network was proposed.First,by combining deep learning with the Bayesian method,a Bayesian neural network based on the probability distribution weight unit was constructed to establish the probability mapping relationship between gas composition and fault state.Then,the Bayesian neural network was trained by varia⁃tional inference to fit the boundary-state transformer’s fault probability and quantitatively evaluate its health using the Monte-Carlo method.Finally,the result of an example shows that the proposed method has a stronger fault diagnosis ca⁃pability for the boundary-state transformer and a good robustness to sample data error.
作者
霍浩
马天龙
李宁瑞
康超
赵立宇
孙伟
HUO Hao;MA Tianlong;LI Ningrui;KANG Chao;ZHAO Liyu;SUN Wei(Extra High Voltage Company,State Grid Gansu Electric Power Company,Lanzhou 730070,China)
出处
《电力系统及其自动化学报》
CSCD
北大核心
2023年第9期137-144,共8页
Proceedings of the CSU-EPSA
关键词
溶解气体分析
变压器
故障诊断
贝叶斯神经网络
变分推理
dissolved gas analysis
transformer
fault diagnosis
Bayesian neural network
variational inference