N-SEMI REGULAR GRAPH-BASED FUZZY SEMI-SUPERVISED LEARNING APPROACH FOR FAULT DETECTION AND CLASSIFICATION IN PHOTOVOLTAIC ARRAYS

Fault identification is a crucial operation for increased performance and reliability of photovoltaic (PV) solar arrays. According to the nonlinear nature of PV, it can be challenging to detect a variety of faults related to conventional protection systems, which contribute to safety concerns and fire risks in PV settings. Machine learning approaches for fault-detection based on parameters like PV arrays, current, irradiance and temperature have been suggested in order to address this safety void. However, current approaches typically utilize supervised learning techniques, which are conditioned by several of the labeled data (so called fault types) and thus have various identified drawbacks such as being costly to implement, model simulation, etc. The paper suggests an N-semi-regular Fuzzy Semi-Supervised Learning System based on an N-semi graph with a few labeled, normalized training data for improved visualization. The proposed model not only describes the loss but also the likely fault structure to facilitate the network recovery. After model building, PV systems can learn to monitor and identify PV failures independently under environmental changes over time. The efficient fault detection and classification of the proposed method show both simulation and experimental results.