On fault detection in a grid-connected photovoltaic system under two different operating modes

Abstract

This project proposes an efficient fault detection method based on process history method using Principal Component Analysis (PCA) technique in dimension reduction and feature extraction for Grid Connected PhotoVoltaic System (GCPV) labeled data. The Fault Indicator (FI) developed in this work is an elliptic threshold based on the gaussian distribution of Normal Operation Condition (NOC) of the two different operating modes, Maximum Power Point Tracking (MPPT) and Power Rating (PR), of the system. The proposed fault detection method has been validated experimentally by the evaluation of the False Alarms Rate (FAR), Fault Detection Rate (FDR) and the Detection Delay (DD).The results shows good performance depending on the selected Confidence Level (CL) 95%, 98% and 99%.