基于LCNN的电弧故障检测方法  

Arc Fault Detection Method Based on LCNN

作　　者：寇皓文 王金龙 KOU Haowen;WANG Jinlong(Faculty of Electrical and Control Engineering,Liaoning Technical University,Huludao 125105,China;Chengde College of Applied Technology,Chengde 067000,China)

机构地区：[1]辽宁工程技术大学电气与控制工程学院,辽宁葫芦岛125105 [2]承德应用技术职业学院,河北承德067000

出　　处：《电工技术》2024年第21期65-71,共7页Electric Engineering

基　　金：2022年度葫芦岛市科技指导计划重点研发项目(编号2022JH2/07b)。

摘　　要：在家用交流供配电系统中,线路老化、接触松动等原因可能会导致电弧故障的发生。电弧故障的危险性极高,可造成严重的电气火灾危害和财产损失。根据家庭负载的实际使用情况,使用了多种不同类型的负载进行串联型电弧故障实验,并获取了不同采样频率下的样本。为了快速而准确地对电弧故障进行检测,使用短时傅里叶变换同时考虑了电流信号的时域和频域特征,将分析结果转换为RGB三色图像作为网络的输入信息。提出了一个轻量型的卷积神经网络LCNN,在网络搭建过程中,同时考虑网络的检测性能和规模,逐步搭建起最优的网络结构。该检测方法具有较好的适应性,能够在5 kHz及以上的采样频率下保持高准确率,并在与其他方法的比较中,证明了其优越的性能。In a household AC power supply and distribution system,arc faults may occur due to reasons such as aging of lines and loose contacts.Arc faults are highly dangerous and can cause serious electrical fires and property damage.In this paper,various types of loads were used in series arc fault experiments based on the actual usage of household loads,and samples were obtained at different sampling frequencies.In order to quickly and accurately detect arc faults,this paper used short-time Fourier transform(STFT) to consider both the time-domain and frequency-domain characteristics of the current signal,and converted the analysis results into RGB color images as the input information of the network.A lightweight convolutional neural network(LCNN) was proposed,and both the detection performance and scale of the network during the network construction process were considered,gradually building up the optimal network structure.This detection method had good adaptability,could maintain high accuracy at sampling frequencies of 5 kHz and above,and in comparison with other methods,was proved to have superior performance.

关 键 词：串联型电弧故障 短时傅里叶变换 卷积神经网络 轻量型 

分 类 号：TM501.2[电气工程—电器]