Fault Classification Using Machine Learning

Abstract

Detecting faults in PV is important for the overall efficiency and reliability of a solar power plant. Ground faults, series and parallel arc faults, high resistance connections, soiling, and partial shadowing need to be detected. The I–V data in a PV array can be measured at the panellevel. This data is useful in predicting possible ground faults or arc faults. The I–V characteristic is a function of temperature, incoming solar irradiance (direct and diffused), angle of incidence, and the spectrum of sunlight. The panel has an optimal operating point for maximum power. Fault detection using I–V data can be accomplished by identifying outliers in the I-V feature space. Current practice is to identify faults via a human operator examining data collected at the inverter. One study identified a Mean Time to Repair (MTTR) of 19 days [6] for a centrally monitored system of residential installations. With the addition of more and higher quality data from SMDs, MTTR could be significantly reduced. Several challenges and research opportunities are evident in the fault diagnosis and localization problems. First, of course, a system must accurately classify the PV array’s condition. It should be able to react to the “unknowns”—faults the system designers did not anticipate. Considering these challenges, several ML approaches can be examined. Simulated fault datawere obtained using the Sandia PV module performance model and a MATLAB circuit simulation package [25, 26].