Supervised Machine Learning and Current Signal Based Universal Fault Protection Strategy for Distributed Generation Integrated into Power Systems

Distributed Generations (DGs) play a significant role in the modern electric power grid, offering benefits such as reduced transmission and distribution losses and minimized environmental impact. Globally, solar Photovoltaic (PV) and Wind Turbine Distributed Generators (WTDGs) have become some of the most widely utilized DG resources, driven by their improving efficiency, and advancing technology. However, from the viewpoint of power system protection, the integration of these resources into power systems poses various significant challenges for protective relays. These relays traditionally rely on both voltage and current measurements and and may not operate correctly in the presence of WTDGs. To address these issues, the use of a current-only protection system with a unified supervised Machine Learning (ML) classifier for directional and phase selection relays is proposed in this work. The comparative analysis shows the superiority of the Random Forest classifier in using only currents and being able to detect both the fault direction with perfect accuracy and the faulty phase(s) with over 92% accuracy, making it suitable for implementation in both transmission and distribution networks, including WTDGs resources.