Non-Intrusive Load Monitoring based Demand Prediction for Smart Meter Attack Detection

Abstract

The global implementation of smart meters that measure and communicate residential electricity consumption has resulted in the creation of new energy efficiency services such as automated energy management systems and billing systems. In view of the vulnerability of smart meters to cyber and physical attacks, this research presents a short-term load prediction method that uses energy disaggregation, to detect the False Data Injection (FDI) attack on smart meters. This method is constructed of an edge detection based Non-Intrusive Load Monitoring (NILM) module for energy disaggregation and a load forecaster. In the first step, we attempt to determine when the appliances are switching on/off. Second, the acquired switching events would be utilized as an input for machine learning algorithms including Support Vector Regression (SVR) and Elman Neural Network (ENN) to improve performance of the load forecaster for detecting FDI attacks. Validation of the results based on the data collected from twenty actual UK houses has indicated that the recommended method is a great solution for detecting cyberattacks on residential smart meters.