FAULT DETECTION IN PHOTOVOLTAIC SYSTEMS USING MACHINE LEARNING APPROACHES

Abstract

The solar photovoltaic (PV) systems are one of the most promising technologies in the generation of environmentally-friendly electricity; however, their overall efficiency is often impaired by the presence of faults disrupting its operation and increasing the maintenance costs1.  Traditional fault-detection methods are based on either manual inspection or unsophisticated threshold models2 which often do not work in a changing environment. We show in the current work that machine-learning algorithms, which are trained on working PV data, are able to effectively differentiate and detect frequent faults in the system. Preprocessing was done in Python and used raw MATLAB data as input, converting it into structured Excel files which were used as training and evaluation inputs. We used a number of classifiers, such as support vegetable machines, random forests, and neural networks, and compared them to each other systematically. The findings reveal that machine-learning models are effective at detecting faults with the best classifier having higher detection rates than 95 percent, thus way outperforming the conventional methods3, 4. These outcomes can make machine-learning a reliable instrument to improve PV systems resiliency. This means that less time is spent offline and this strategy minimizes maintenance bills, as well as helping the wider adoption of reliable and large-scale solar energy infrastructure.