基于自编码器和改进K均值聚类的光伏组件故障诊断

Fault Diagnosis of Photovoltaic Modules Based on Auto-Encoder and Improved K-Mean Clustering

光伏组件运行时关键参数具有连续性,难以直接用聚类的方式诊断故障,因此提出了基于自编码器(Autoencoder,AE)和改进K均值聚类K-Means++光伏组件故障诊断方法。利用AE降维数据输入特征数据,保证了特征之间的高内聚、低耦合以及去线性的特性。针对传统K均值聚类(K Means Clustering,K-Means)对初始聚类中心敏感,且聚类结果随不同的初始聚类中心波动的缺陷,基于数据分布特征选取初始聚类中心的K-Means++算法。数据试验结果表明,基于AE和K-Means++算法对短路、开路、阴影遮挡以及老化故障的诊断表现良好。

Based on the continuity of the key parameters in the running time of photovoltaic modules,it is difficult to directly use clustering method for fault diagnosis.Therefore,a photovoltaic module fault diagnosis method based on Autoencoder(AE)and improved K-Means++clustering is proposed.AE was used to reduce the dimension of data input feature data to ensure high cohesion,low coupling and linear characteristics between features.Aiming at the defect that traditional K Means Clustering(K-Means)is sensitive to the initial clustering center and the clustering results fluctuate with different initial clustering centers,the K-Means++algorithm based on data distribution features is adopted to select the initial clustering center.The data test results show that the algorithm based on AE and K-Means++performs well in fault diagnosis of short circuit,open circuit,shadow occlusion and aging.