Detection of non-technical losses using advanced metering infrastructure and deep recurrent neural networks

Abstract

Non-Technical Losses including power theft is a major problem faced by power utilities today. In India, this accounts for roughly 50% of the total loss every year. Non-Technical losses can be infidelity at the consumer end, unethical tapping of transmission lines and hacking or tampering of energy meters. In this paper, a solution to power theft using advanced metering infrastructure and intelligent algorithms has been proposed. The proposed solution profiles users based on their half hour wise power consumption data and locality. Irregularities in power usages are detected using a state-of-the-art artificially intelligent algorithm: Recurrent Neural Networks. The algorithm uses Long Short Term Memory (LSTM) units to process sequential power consumption data. The result will be a model that can be used to shortlist consumers who are potentially stealing power, in real time.