PV System Fault Classification using SVM Accelerated by Dimension Reduction using PCA

Abstract

Fault diagnosis of photovoltaic systems is imperative for cost-effectiveness and improving the efficacy of the plant. Owing to the non-linear output characteristics of the PV arrays, the detection of faults becomes tough for conventional protective equipment. The paper presents a detailed procedure for the diagnosis of four kinds of faults in the PV system, namely the short circuit fault, inverse bypass diode fault, shunted bypass diode fault, and shadowing effect in modules using Support Vector Machines (SVM) based classification strategy. Principal Component Analysis (PCA) helps in the feature extraction and projection of original data into lower-dimensional space. The SVM aided by PCA, addressing the humungous multi-class problem, is compared with other machine learning strategies like the Feed Forward Neural Network using the Levenberg-Marquardt damped Newton backpropagation technique, and exact fit Radial Basis Functions. The proposed model uses synthetic data from the PVLIB toolbox. The decision boundaries are toggled along specified vectors to check the effect on the fault confusion matrix. The simulated results signify that the fault diagnosis schema can correctly classify faults with high efficiency.