基于轻量型CNN和特征阈值的光伏系统串联电弧故障检测装置  

Design of Photovoltaic Series Arc Fault Detection Device Based on Lightweight CNN and Feature Threshold

作　　者：王兆锐 何键涛 李治彤 鲍光海[1] WANG Zhaorui;HE Jiantao;LI Zhitong;BAO Guanghai(College of Electrical Engineering and Automation,Fuzhou University,Fuzhou 350108,China;College of Physics and Information Engineering,College of Microelectronics,Fuzhou University,Fuzhou 350108,China)

机构地区：[1]福州大学电气工程与自动化学院,福建福州350108 [2]福州大学物理与信息工程学院、微电子学院,福建福州350108

出　　处：《电器与能效管理技术》2024年第8期77-85,共9页Electrical & Energy Management Technology

基　　金：福建省科技计划资助项目(2023H0007)。

摘　　要：为保证光伏发电系统的安全稳定运行,提出一种基于轻量型卷积神经网络(CNN)和特征阈值的光伏串联电弧故障检测算法。为了应对逆变器异常工况和光伏阵列时变性对信号特征的影响,以及不同弧长(0.05~10.00 mm)导致的信号特征差异,利用高频耦合信号为特征信号,并结合神经网络算法和特征阈值方法,检测光伏线路上的串联电弧故障。最后,制作光伏串联电弧故障检测装置样机。经实验测试,样机切断电弧故障的平均时间为177.1 ms,且在逆变器异常工况的干扰下不会发生误判。To ensure the safe and stable operation of photovoltaic(PV)systems,a PV series arc fault detection algorithm based on a lightweight convolutional neural network combined with feature threshold methods is proposed.To addresses the impacts of inverter abnormal conditions and the time-varying nature of PV arrays on signal characteristics,as well as the signal characteristic differences caused by varying arc lengths(0.05~10.00 mm),by utilizing high-frequency coupled signals as feature signals and combining neural network algorithms with feature threshold methods,the algorithm detects series arc faults in PV circuits.Finally,a prototype of a PV series arc fault detection device is created.The experimental tests show that the prototype cuts off arc faults in an average time of 177.1 ms and does not produce false positives under the inverter abnormal conditions.

关 键 词：光伏系统 串联电弧故障 卷积神经网络 电弧检测装置 

分 类 号：TM501.2[电气工程—电器]