Machine Learning for Cyber-Physical Power System Security

Abstract

The rapidly growing deployment of cyber components in modern power systems increases the vulnerability to cyberattacks, which significantly impact the security of energy usage and can potentially induce large-scale blackouts in extreme scenarios. Although the defending techniques have been intensively studied in existing works, most of them highly rely on the physical models and explicit mathematical formulations to identify the abnormalities. However, these methods are usually difficult to apply in the cyber-physical power system (CPPS) due to the high complexity and uncertainty, thereby resulting in inferior scalability and accuracy. To address these problems, machine learning (ML) techniques can provide alternative approaches to analyze the security and safety issues in CPPS and tackle the potential threats more effectively. The experimental simulations have been implemented to verify the effectiveness of these techniques.