Anomaly Detection in Smart Grids using Machine Learning Techniques

Abstract

Monitoring a smart grid using Information and communication technology tools require mechanisms for detection of anomalies like faults, cyber-attacks, etc. Most of the present power systems are equipped to deal with anomalies caused by faults or other natural disturbances. However, detecting and discriminating anomalies caused by the cyberattack against the power system is yet to be satisfactorily achieved. In this paper, we employ different machine learning techniques to detect power system anomalies. Besides detecting anomalies, machine learning methods can also differentiate between cyberattacks and natural disturbances. We assess several states of the art machine learning techniques in terms of their ability to detect cyber-attacks including those attacks which employ deceptive techniques to hide their foot prints. Evaluation is carried out using a dataset generated by an IEEE 3-bus system, and performances of machine learning techniques are analyzed in terms of accuracy of detection and required processing power.