Fault detection and classification in a DG powered LVDC distribution system using machine learning algorithm

Abstract

Distributed Generation has become an integral part of the microgrid system predominantly powered by solar PV systems. Electric Vehicles, renewable energy sources, and household appliances are just a few examples of the increasing number of DC loads that are driving the growing significance of Low-Voltage DC distribution networks. Higher power transfer capacity than AC, lower energy conversion losses, and increased efficiency and dependability are some benefits of low voltage DC systems. It has become very essential that faults that occur in such systems must be detected and the type of fault must be identified accurately so that the system’s reliability can be further increased. The literature provides many methodologies for identifying and classifying the faults in AC transmission systems and also in LVDC distribution systems. In off grid LVDC distribution systems, approaches such as deep learning based identification and classification of faults is presented in literature, which majorly concentrates on small electrification and poor internet coverage area of Sub-Saharan Africa. A methodology based on power electronic converter is also presented in literature for fault diagnosis, this methodology includes signal injection, which may lead to line interferences. To overcome these challenges, this paper proposes a new methodology for identifying and classifying the faults in renewable based LVDC distribution systems using machine learning algorithms such as k-Nearest Neighbour (kNN) and Decision Tree (DT). Literature presents a maximum of 99 % of accuracy in identifying and classifying the faults whereas, the proposed methodology achieves 100 % accuracy in identifying and classifying the faults in LVDC distribution system with 100 % precision.