基于伪标签-1D DenseNet-KNN的光伏阵列开集复合故障诊断方法  

Photovoltaic array open set compound fault diagnosis based on pseudo-label-1D DenseNet-KNN

作　　者：陈泽理 卢箫扬 林培杰 赖云锋 程树英 陈志聪 吴丽君 CHEN Zeli;LU Xiaoyang;LIN Peijie;LAI Yunfeng;CHENG Shuying;CHEN Zhicong;WU Lijun(Institute of Micro-Nano Devices and Solar Cells,College of Physics and Information Engineering,Fuzhou University,Fuzhou,Fujian 350108,China)

机构地区：[1]福州大学物理与信息工程学院,微纳器件与太阳能电池研究所,福建福州350108

出　　处：《福州大学学报（自然科学版）》2023年第4期490-497,共8页Journal of Fuzhou University(Natural Science Edition)

基　　金：福建省自然科学基金面上资助项目(2021J01580);福建省科技厅引导性基金资助项目(2022H0008);福建省工信厅资助项目(82318075);福州市科技计划资助项目(2021-P-030)。

摘　　要：提出一种基于伪标签-1D DenseNet-KNN的光伏阵列故障诊断方法,实现在少标签样本下的光伏阵列复合故障开集识别.首先,分析各种常见单一故障和灰尘覆盖下复合故障的I-V特性曲线;然后,为克服常规半监督机器学习算法需手动提取数据特征的问题,采用一种伪标签与1D DenseNet相结合的半监督方法自动提取特征;最后,将从训练数据中提取的特征、训练数据预测的标签和测试样本提取的特征输入KNN算法并进行开集复合故障诊断.实验表明,该方法不仅能准确分类各种已知类别样本,还能识别出未知类别故障,且模型训练只需要少量的标签数据.A fault diagnosis method based on pseudo-label-1D DenseNet-KNN with fewer labeled samples is proposed to classify open-set of compound faults for photovoltaic arrays.First,the I-V characteristic curves of various common single faults and dust covered compound faults are analyzed.Since the semi-supervised machine learning algorithms need to manually extract some features from the data,then a semi-supervised method combining pseudo-labels with 1D DenseNet is used to automatically extract features.Finally,the features extracted from the training data,the predicted labels of the training data and the features extracted from the test samples are inputted into the KNN algorithm for open set compound fault diagnosis.Experiments show that this method can not only classify various known class samples accurately,but also identify unknown class faults,and the training of the model only needs a small amount of labeled data.

关 键 词：光伏阵列 故障诊断 I-V特性曲线 伪标签半监督学习 开集识别 KNN算法 

分 类 号：TM615[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程]