Feature Selection based False Data Detection Scheme using Machine Learning for Power System

Abstract

An electrical power grid is a conglomerate system that requires meticulous monitoring to ensure uninterrupted, secured and reliable grid operation by incorporating state estimation to ensure a better estimate of the power grid state through assessment of meter quantification. The state estimator operates on real-time inputs that are data and status information. Thereby, it becomes necessary to automatize and digitize the electric grid by enhancing the widespread installation of Remote Terminal (RTUs) and Phasor Measurement Units (PMUs) for improvising real-time wide-area system monitoring and control. However, the challenge of anomaly detection of the data obtained from the PMUs still exists as the PMUs data comprises different types of anomalies arising from both physical and cyber systems. This work proposes a machine learning-based scheme to detect the anomaly in the data. Principal Component Analysis algorithm is used as the feature selection algorithm to attain the important characteristics in the data and then a supervised classification algorithm is used to obtain the erroneous data in the PMU data streams.