Data-driven Energy Theft Detection in Modern Power Grids

Abstract

Energy theft is an old and multifaceted phenomenon affecting our society on a global scale from both an operational as well as from a monetary perspective. The relatively recent decentralisation of the grid infrastructure with the integration of Distributed Renewable Energy Resources (DRES) in synergy with the widely adopted demand-response business model, has undoubtedly broadened the spectrum of attack surface enabling energy theft. Conventional data-driven energy theft detection schemes have a strong dependency on assessing the spatio-temporal patterns of SCADA measurements aggregated at the Distribution System Operator (DSO) or Transmission System Operator (TSO) with minimal consideration of the intrinsic weather patterns related to individual DRES deployments. Hence, theft scenarios instrumented by DRES owners consuming the energy they produce (i.e., prosumers) can effectively be stealthy and hard to spot. Therefore, in this work we introduce a data-driven, SCADA-agnostic energy theft detection framework explicit to DRES-based scenarios. We provide a comprehensive formalisation of a DRES-based theft attack model and further assess the performance of our framework by utilising and relating freely available third-party weather measurements with real solar and wind turbine deployments in Australia and France. Evidently, our proposed framework yields an energy theft detection accuracy rate of over 98% with optimal computational costs. Thus, reasonably addressing the highly demanding requirements of low-cost and accurate real-time energy theft detection in modern power grids.