Efficient Fault Detection and Diagnosis Procedure for Solar Photovoltaic Array Based on Multiclass Support Vector Machine

Abstract

Solar photovoltaic systems installed in outdoor environments are susceptible to faults and partial shading, which leads to reduction in the maximum power generated. Fire risks and decreased system effectiveness emerge from the standard protection devices’ inability to identify line-line faults due to their non-linear properties. In this paper, a Multiclass Support Vector Machine (MSVM)-based fault identification algorithm is proposed to detect line-ground (LG), line-line (LL) faults and partial shading (PS). The P–V, I–V characteristics and the string current waveforms are analysed for LG, LL faults and PS conditions. Two features, impedance and alpha, are computed to detect the fault conditions. The Multiclass SVM-based classification algorithm is used to classify the dataset using One vs Rest classifier. The three different kernel functions are used to train the SVM that include linear, sigmoid and radial-basis function kernels. The performance of the proposed Multiclass SVM algorithm is experimentally tested using a 1.6 kW, 44 photovoltaic solar PV array.