基于深度学习的电容器介损角在线辨识  

Online Identification Method of Power Capacitor Dielectric Loss Angle Based on Deep Learning

作　　者：王晓辉[1] 朱永利[1] 王艳[1] 郭丰娟[1] 

机构地区：[1]华北电力大学控制与计算机工程学院,保定071003

出　　处：《电工技术学报》2017年第15期145-152,共8页Transactions of China Electrotechnical Society

基　　金：国家自然科学基金(51407076;51677072);中央高校基本科研业务专项资金(2014MS131)资助项目

摘　　要：当前电容器介质损耗因素的计算方法为正向求解过程,即先对电容器工作电流和电压进行采样,再使用谐波分析等方法计算介损值,实践中算法稳定性不佳。为此提出了一种基于深度学习的电容器介损角辨识方法,采用一段时间的监测值训练深度学习网络,再使用该深度学习网络对新采样的信号进行辨识,判断介损角变化量(分辨率为0.001%)。给出了用于深度学习的介损角表示信号Dδ(t)的计算过程,证明了在讨论域内该信号的幅值即是介损角δ,且其波形形状包含监测装置受到的干扰。仿真实验证明该方法有效,比加汉宁窗的谐波分析法具有更好的抗噪能力。实际在线监测样本的计算结果表明其稳定性优于加汉宁窗的谐波分析法,且辨识结果不受电压互感器角差的影响。Most of the algorithm of calculating dielectric loss factor are positive solving process, which include sampling capacitor current and voltage, and calculating the dielectric loss factor of these signals by harmonic analysis. These methods have poor robustness when there are unidentified distortions in the sampling signal. This paper proposes a capacitor dielectric loss factor identification algorithm based on the deep learning. The algorithm proposed in this paper trains a feed-forward multilayer artificial neural network with a period of online sampling signals, and identify the dielectric loss angle from new monitoring data with resolution of 0.001%. The computation of dielectric loss factor identification signal Ds （ t ） is proposed, and verify the amplitude of Ds（ t ） is the dielectric loss angle. And the shape of its waveform includes the interference of the monitoring device. The validity of the method has been proved by simulation. The method can achieve better ability to resist noise than hanning windows harmonic analysis method. The calculation results based on actual online monitoring data shows better robustness than hanning windows harmonic analysis method, its results is also not affected by the angle error of potential transformer.

关 键 词：深度学习 电容器 介损角 在线监测 

分 类 号：TM835.4[电气工程—高电压与绝缘技术]