摘要
准确诊断金属氧化物压敏电阻故障对电力系统安全运行十分重要。为有效提高故障诊断率,本文提出基于主成分分析与反向传播神经网络相结合的故障诊断算法。对于MOV故障数据指示指标,利用主成分分析进行降维,降低原始指标间相关性,最后利用反向传播神经网络对所选主成分进行诊断。试验结果表明:相对于反向传播模型,PCA-反向传播模型诊断误判率下降了3%,模型运行时间降低了267 s。有效降低MOV误判率,简化了反向传播网络结构,提高网络收敛性和稳定性。
Accurate fault diagnosis of the metal oxide varister is very important for the safe operation of power system. In order to improve the MOV fault diagnosis rate,the PCA-BP algorithm is proposed.The principal component analysis is used to reduce the dimension of MOV diagnosis indexes and the correlation between the original indexes. The new indexes are input to the BP neural network to train,so the MOV fault diagnosis can be achieved. The results show that,compared with BP model,the diagnosis error rate of PCA-BP model decreased by 3%,and the running time of this model decreased by 267 s. It effectively reduces the MOV misjudgment rate,simplifies the BP network structure,and improves the convergence and stability of the network.
作者
王伟
虢韬
杨恒
WANG Wei;GUO Tao;YANG Heng(Power Grid Planning Research Center,Guizhou Power Grid Co.,Ltd.,Guiyang 550002,China;Transmission Operation and Maintenance Branch,Guizhou Power Grid Co.,Ltd.,Guiyang 550002,China)
出处
《电瓷避雷器》
CAS
北大核心
2019年第6期20-25,共6页
Insulators and Surge Arresters
基金
北斗卫星电力行业应用关键技术研究及数据应用产业化(编号:黔科合重大专项字【2018】3007).
作者简介
王伟(1986—),男,本科,研究方向为输电线路设计、评审及运行维护。
