Conventional KPCA Approach Applied to Detect Simulated Faults in PV Systems Using Simulated Data

Abstract

Photovoltaic (PV) installations have become integral for harnessing solar energy, yet ensuring uninterrupted power generation remains crucial. This study addresses the challenge of maintaining reliability in PV systems by proposing a method to detect and identify simultaneous faults, using kernel principal component analysis (KPCA) and statistical metrics. The proposed method employs KPCA, a machine learning technique adept at identifying patterns in complex data. By utilizing statistical metrics in a feature space generated by KPCA, potential faults in PV system performance data are flagged. Unlike prior research that focused on single faults, this work extends the application of KPCA to detect and identify multiple faults occurring simultaneously, such as partial shading combined with open or short circuit faults. Through extensive simulations, including 100 samples of different faults under varying irradiance conditions, the method demonstrates high accuracy rates: 93.33% for partial shading, 100% for open circuit, 100% for short circuit, and 81.81% for combinations of partial shading with either open or short circuit faults. Results from a Matlab‐Simulink model validate the effectiveness of KPCA in detecting both single and simultaneous faults in PV systems’ DC side.