Fault Classification with Robust Knowledge Transfer for Single Phase Grid Connected Photovoltaic Systems

Disturbances in grid connected photovoltaic systems (GCPVS) can cause voltage collapse, trigger shutdown of the power electronic converters and cause voltage fluctuations in the grid. These faults may result in harmonics, asymmetry, flicker and stability problems. Hence, continuous monitoring of GCPVS to classify the fault or operating condition of the system is necessary. In this paper, a fault classification method for GCPVS is developed using a modified method of knowledge transfer incorporating cross-validation methods. The adapted method learns the system characteristics using privileged information in the training set. The proposed classification method is simulated by considering eight different GCPV faults and subjecting them to cepstrum feature extraction. The results depicted 98.5% classification accuracy and 1.5% error rate, which is better when compared with the conventional classification process.