A New Theft Detection Approach for Cyberattacks in PV Generation

Smart meters on customer premises are used to measure their consumption as well as generation in the case of installed local generation units. The utility then compensates the customers for the amount of electricity they generate. Manipulating smart meters readings to give false electricity consumption or generation is becoming more common, putting the utilities' revenues at risk. Unlike most of the work in this area, this paper deals with theft in the generation domain, not the consumption domain. In specific, the objective of this work is to apply regression machine learning to detect electricity theft in photovoltaic (PV) generation. We propose a new theft detection approach based on a regression tree model to identify suspicious information. The proposed approach relies on a combination of machine learning and statistical analysis, where the error between the estimated PV output and the reported PV output is used to classify the customer. If the error is highly unlikely to happen, then the customer is classified as malicious; otherwise, the customer is honest. For testing and training, we use 25 different panels, where 24 panels are used for training and the 25th panel is used for testing. Simulation results prove the effectiveness of the proposed approach.