An effective DNN-based Approach for Detecting Energy Theft in Smart Grids through User Consumption Patterns

Abstract

The advancement of the Internet has been progressively easing human life. The development of mobile communication technologies has led to the widespread adoption of Internet of Things (IoT) applications. Thus, most systems and devices have connected to the Internet more efficiently. The integration of communication systems into critical infrastructures, such as electricity grids, has given rise to the concept of IoT-based smart grids. In smart grid systems, data communication is facilitated through the Advanced Metering Infrastructure (AMI). Due to the inherent characteristics of communication systems, AMI may be vulnerable to cyber-attacks. Some vulnerabilities have resulted in the emergence of cyber-attack vectors against energy consumption data obtained from smart meters. In this study, an effective energy theft intrusion detection system (IDS) based on users' consumption patterns is proposed. A Deep Neural Network (DNN) based classification model was employed to assess the predictability of both honest and malicious consumption patterns. The proposed model exhibits high and adjustable performance. Extensive experiments have been carried out on a real consumption dataset of approximately 2000 customers. Manipulated data from real readings with two different attack vectors were injected into the dataset. K-fold cross-validation technique was used. The proposed model performed a high performance reaching up to 97.4% accuracy.