Detecting Energy Theft and Anomalous Power Usage in Smart Meter Data

Abstract

The success of renewable energy usage is fuelling the power grids most significant transformation seen in decades, from a centrally controlled electricity supply towards an intelligent, decentralized infrastructure. However, as power grid components become more connected, they also become more vulnerable to cyber attacks, fraud, and software failures. Many recent developments focus on cyber-physical security, such as physical tampering detection, as well as traditional information security solutions, such as encryption, which cannot cover the entire challenge of cyber threats, as digital electricity meters can be vulnerable to software flaws and hardware malfunctions. With the digitalization of electricity meters, many previously solved security problems, such as electricity theft, are reintroduced as IT related challenges which require modern detection schemes based on data analysis, machine learning and forecasting. The rapid advancements in statistical methods, akin to machine learning techniques, resulted in a boosted interest towards concepts to model, forecast or extract load information, as provided by a smart meter, and detect tampering early on. Anomaly Detection Systems discovers tampering methods by analysing statistical deviations from a defined normal behaviour and is commonly accepted as an appropriate technique to uncover yet unknown patterns of misuse. This work proposes anomaly detection approaches, using the power measurements, for the early detection of tampered with electricity meters. Algorithms based on time series prediction and probabilistic models with detection rates above 90% were implemented and evaluated using various parameters. The contributions include the assessment of different dimensions of available data, introduction of metrics and aggregation methods to optimize the detection of specific pattern, and examination of sophisticated threads such as mimicking behaviour. The work contributes to the understanding of significant characteristics and normal behaviour of electric load data as well as evidence for tampering and especially energy theft.

Date of Award	2020
Original language	English
Awarding Institution	University of Plymouth
Supervisor	Martin Kappes (Other Supervisor)