基于CONMI特征选择和支持向量机的光伏系统并发故障精确诊断

Abstract Photovoltaic (PV) systems are susceptible to faults in outdoor environment. Multiple faults can exist in PV systems simultaneously, the phenomenon known as concurrent faults. An accurate identification of concurrent faults is critical as they pose safety risks and lead to financial losses. More often than not, these faults exhibit highly identical electrical features which cannot be differentiated using conventional methods. While single fault diagnosis in PV systems has been extensively studied, the research on concurrent faults is lacking. In this study, a wide range of concurrent faults datasets are considered and analysed using key features selected via filter feature selection algorithm, CONMI. To analyze manageable number of concurrent faults, this study focuses on dual-fault combinations. From the findings, Support Vector Machine (SVM)—chosen for its simplicity and effectiveness in distinguishing high-dimensional dataset—is applied for the fault identification and diagnosis task. Additionally, the suitability of several SVM kernels, namely, linear, polynomial, radial basis function, and sigmoid, are compared and discussed. To validate CONMI, classification results of SVM models using CONMI-selected features are compared to Principal Component Analysis and Normalized Mutual Information. Additionally, an increase in computational efficiency of SVM models is determined through comparison of model computation times with and without CONMI-selected features. For further validation of SVM models, their results are benchmarked against other widely used machine learning methods i.e., Random Forest, Logistic Regression, and k -Nearest Neighbors. The results show that CONMI reduces computation time by up to 18 times while polynomial kernel-based SVM model outperformed its competitors, achieving an overall classification accuracy of 94.52%. Further, it was found that concurrent faults involving line-to-line faults and partial shading conditions are particularly challenging to diagnose, irrespective of the environmental conditions, due to their almost indistinguishable variations in the I-V characteristics.