基于改进深度残差收缩网络的光伏发电阵列故障诊断方法  

Fault diagnosis method of microgrid system based on improved deep residual shrinkage network

作　　者：彭辉[1] 黄婧柠 郑宇锋 田程程 严路 PENG Hui;HUANG Jingning;ZHENG Yufeng;TIAN Chengcheng;YAN Lu(School of Electrical Engineering and Automation,Wuhan Univ.,Wuhan 430072,China;National Key Laboratory of Electromagnetic Energy,Naval Univ.of Engineering,Wuhan 430033,China;Unit No.92011,Shanghai 201913,China)

机构地区：[1]武汉大学电气与自动化学院,武汉430072 [2]海军工程大学电磁能技术全国重点实验室,武汉430033 [3]92011部队,上海201913

出　　处：《海军工程大学学报》2024年第6期1-8,共8页Journal of Naval University of Engineering

基　　金：国家自然科学基金资助项目(92266201);中国博士后科学基金资助项目(2019T120972)。

摘　　要：光伏发电存在诸如出力随机性强,受气象和环境因素影响大,易受接线方式和光伏电池组件内部健康状态等影响的问题。针对上述问题,将光伏阵列输出电压及各支路输出电流波形图像作为故障诊断模型的输入,并对深度学习典型算法中的卷积神经网络和深度残差网络进行改进,以适用于二维图像类型辨识且特征提取性能更佳的深度残差收缩网络作为光伏阵列故障诊断算法,在Matlab/Simulink中建立并网光伏发电系统仿真模型,并搭建了与之相对应的试验平台,分别测量正常运行及各类故障下的光伏阵列输出电压以及各支路输出电流,并绘制相应波形特征图作为深度残差收缩网络算法的输入样本,实现并网光伏阵列的故障分类辨识。数值仿真与试验验证了深度残差收缩网络模型的正确性与优越性,对比分析结果表明:该算法在并网光伏阵列故障诊断仿真中的准确率显著高于卷积神经网络和残差网络算法,具有更佳的训练效果和分类性能。Photovoltaic power generation may encounter such problems as randomness of output,being easily affected by meteorological and environmental factors,and aptness to be subjected to wiring mode and the internal health status of photovoltaic cell modules.Aiming at solving the above prob-lems,the waveform images of the photovoltaic array output voltage and the output current of each branch were taken as the input of the fault diagnosis model.Furthermore,CNN and DRN in typical deep learning algorithms were improved.DRSN was suitable for 2D image type identification and has better feature extraction performance,which was used as the PV array fault diagnosis algorithm.The simulation model of grid-connected photovoltaic power generation system was built in Matlab/Simu-link,and the corresponding test platform was built.The output voltage of photovoltaic array and out-put current of each branch under normal operation and various faults were measured respectively,and the corresponding waveform characteristic map was drawn as the input sample of DRSN algorithm.Thus the fault classification and identification of grid-connected photovoltaic array are realized.Nu-merical simulation and experiments verify the correctness and superiority of DRSN model.The results of comparative analysis show that the accuracy of DRSN algorithm in grid-connected photovoltaic ar-ray fault diagnosis simulation is significantly higher than that of CNN and ResNet algorithm,and thus the algorithm has better training effect and classification performance.

关 键 词：深度残差收缩网络 故障诊断 并网光伏发电系统 波形特征图 

分 类 号：TM914[电气工程—电力电子与电力传动]