Cyber-Attack Detection in Smart Grids Using Machine Learning Approach

To meet the increasing demand of reliable energy, now-a-days power grids rely on synchronous phasor systems, Advanced Metering infrastructure (AMI) and industrial automation control systems. The critical cyber physical infrastructure and high-speed networks in smart grid has made it susceptible to cyber-attacks. As a centrally controlled infrastructure smart grid captures and processes data with the help of several devices such as smart meters, sensors and virtual machines. Such a huge amount of data if not handled properly can trigger massive blackouts and infrastructural damage to the smart grid. To assess the state of the power system using conventional methods, phasor quantity measurements are used. The observed and estimated findings are compared with a limit for cyber-attack detection. However, modern smart grids, which are mostly built on supervised learning algorithms, frequently use machine learning (ML) to address cyber security challenges. In this context, the main focus of this project is to differentiate between natural fault, cyber-attack and no fault or normal operating condition. The results of various ML algorithms are compared and best possible model is suggested as an enhancement to existing available methods.