基于卷积神经网络的高压输电线路故障定位时域法  

Time-domain Fault Location Method for High-voltage Transmission Line Based on Convolutional Neural Network

作　　者：滕志鹏 梁远升[1] 曾德辉 黄凤丽 TENG Zhipeng;LIANG Yuansheng;ZENG Dehui;HUANG Fengli(College of Electric Power,South China University of Technology,Guangzhou,Guangdong 510641,China;Guangzhou Jiayuan Electric Power Technology Co.,Ltd.,Guangzhou,Guangdong 510610,China;Guangxi Power Grid Co.,Ltd.,Nanning,Guangxi 530023,China)

机构地区：[1]华南理工大学电力学院,广东广州510641 [2]广州嘉缘电力科技有限公司,广东广州510610 [3]广西电网有限责任公司,广西南宁530023

出　　处：《广东电力》2021年第6期1-9,共9页Guangdong Electric Power

基　　金：广东省基础与应用基础研究基金项目(2021A1515010631);广东省自然科学基金项目(2018A030310354)。

摘　　要：受仿真模型参数与实际参数差异的影响,现有基于深度学习的故障定位方法所构建的模型对线路参数较为敏感,制约了该类方法的推广应用。为此,将输电线路的Bergeron模型和卷积神经网络(convolutional neural network,CNN)算法相结合,建立基于CNN的沿线补偿电压波形相似度评估模型,进而提出一种输电线路双端故障定位时域法。首先对线路两端电压、电流量进行相模变换,利用输电线路Bergeron模型从线路两端推算沿线补偿电压波形,利用多组补偿电压波形训练样本对CNN模型进行训练;然后,采用训练好的CNN模型评估线路沿线各观测点补偿电压波形相似度,以故障点波形相似度最高为依据进行故障定位;最后,应用PSCAD进行故障仿真,对所提方法进行验证。仿真结果表明:所构建的波形相似度CNN模型对线路参数不敏感;对于不同参数的线路,所提方法无需重新对CNN模型进行训练就可以准确定位故障,有利于该方法的实际应用。Affected by the difference between simulation model parameters and actual parameters,the model built by the fault location method based on existing deep learning method is sensitive to transmission line parameters,which restricts the popularization and application of these methods.For this reason,this paper combines the Bergeron model of the transmission line with the convolutional neural network(CNN)algorithm to establish a CNN-based similarity evaluation model of the compensation voltage waveform along the line and then proposes a time-domain fault location method for the transmission line.This method firstly performs phase-mode conversion on the voltage and current at both ends of the line,adopts the Bergeron model of the transmission line to calculate the compensation voltage waveform along the line from both ends,and uses multiple sets of training samples to train the CNN model.Afterwards,it used the trained CNN model to evaluate the similarity of the compensation voltage at each observation point along the line,and locate the fault based on the highest similarity evaluation.Finally,the paper uses the PSCAD model for fault simulation to verify the proposed method.The results show that the established CNN model for waveform similarity is not sensitive to transmission line parameters,and this method can accurately locate fault for the lines with different parameters without retraining the CNN model,which is beneficial to the practical application of this method.

关 键 词：卷积神经网络 波形相似度 输电线路 故障定位 

分 类 号：TM855.2[电气工程—高电压与绝缘技术] TP183[自动化与计算机技术—控制理论与控制工程]