Fault detection in power grids based on improved supervised machine learning binary classification

With the increased complexity of power systems and the high integration of smart meters, advanced sensors, and high-level communication infrastructures within the modern power grids, the collected data becomes enormous and requires fast computation and outstanding analyzing methods under normal conditions. However, under abnormal conditions such as faults, the challenges dramatically increase. Such faults require timely and accurate fault detection, identification, and location approaches for guaranteeing their desired performance. This paper proposes two machine learning approaches based on the binary classification to improve the process of fault detection in smart grids. Besides, it presents four machine learning models trained and tested on real and modern fault detection data set designed by the Technical University of Ostrava. Many evaluation measures are applied to test and compare these approaches and models. Moreover, receiver operating characteristic curves are utilized to prove the applicability and validity of the proposed approaches. Finally, the proposed models are compared to previous studies to confirm their superiority.