Graph-based semi-supervised learning for fault detection and classification in solar photovoltaic arrays

Abstract

Fault detection in solar photovoltaic (PV) arrays is an essential task for increasing reliability and safety in PV systems. Fault classification allows identification of the possible fault type so that to expedite PV system recovery. However, because of the non-linear output characteristics of PV arrays, a variety of faults may be difficult to detect using conventional protection devices. Supervised learning methods have been previously proposed to detect and classify solar PV arrays. These methods rely on numerous labeled data for training models and, therefore, have drawbacks: 1) The labeled data on solar PV arrays is difficult or expensive to obtain; 2) The model requires updates as environmental conditions change. To solve these issues, this paper proposes a fault detection and classification method using graph-based semi-supervised learning (SSL). The proposed method only uses a few labeled data points, but relies instead on a large amount of inexpensive unlabeled data points. The method demonstrates self-learning ability in real-time operation. Simulation and experimental results verify the proposed method.