Electricity theft detection framework based on universal prediction algorithm

Abstract

Electricity theft has caused huge losses over the globe and the trend of its perpetuation constantly evolve even as smart technologies such as smart meters are being deployed. Although the smart meters have come under some attacks, they provide sufficient data which can be analysed by an intelligent strategy for effective monitoring and detection of compromised situations. So many techniques have been employed but satisfactory result is yet to be obtained for a real-time detection of this electrical fraud. This work suggests a framework based on Universal Anomaly Detection (UAD) utilizing Lempel-Ziv universal compression algorithm, aimed at achieving a real-time detection in a smart grid environment. A number of the network parameters can be monitored to detect anomalies, but this framework monitors the energy consumption data, rate of change of the energy consumption data, its date stamp and time signatures. To classify the data based on normal and abnormal behaviour, Lempel-Ziv algorithm is used to assign probability of occurrence to the compressed data of the monitored parameters. This framework can learn normal behaviours of smart meter data and give alerts during any detected anomaly based on deviation from this probability. A forced aggressivemeasure is also suggested in the framework as means of applying fines to fraudulent customers.