基于卷积神经网络多判据融合的井下电网故障选线方法  

Fault Line Selection Method for Mine Power Grid Based on Fusion of Multiple Criteria of Convolutional Neural Network

作　　者：王玉梅[1] 张家康 WANG Yumei;ZHANG Jiakang(School of Electrical Engineering and Automation,Henan Polytechnic University,Jiaozuo 454000,China)

机构地区：[1]河南理工大学电气工程与自动化学院,焦作454000

出　　处：《电源学报》2023年第5期110-117,共8页Journal of Power Supply

基　　金：河南省科技攻关资助项目(112102210004)。

摘　　要：针对矿井电网消弧线圈接地系统单相接地故障选线方法准确率和可靠性不高的问题,提出基于卷积神经网络多判据融合的选线方法。主要分析了深度学习模型——卷积神经网络的结构与原理,通过快速傅里叶变换和小波变换从故障信息中提取5次谐波分量、小波分析模极大值、衰减直流分量和高频暂态分量作为原始输入数据,并利用改进LeNet-5模型强大的学习能力和泛化能力对其进行融合。基于Matlab软件搭建井下电网仿真模型,结果表明该方法准确性高、可靠性强。To solve the problem of low accuracy and low reliability in the single-phase grounding fault line selec-tion method for an arc suppression coil earth system in mine power grid,a line selection method based on the fusion of multiple criteria of convolutional neural network(CNN)was proposed.The structure and principle of the deep learning model(i.e.,CNN)were analyzed.Through fast Fourier transform and wavelet transform,the fifth harmonic component,wavelet transform modulus maximum,decaying DC component and high-frequency transient component were extracted from the fault information and taken as the original input data,which were further fused by means of the powerful learning and generalization capabilities of an improved LeNet-5 model.A simulation model of mine power grid was built based on Matlab.Results shown that the proposed method is accurate and reliable.

关 键 词：深度学习 卷积神经网络 故障选线 矿井供电 信息融合 

分 类 号：TM773[电气工程—电力系统及自动化]