Edge ML and IoT-Based Real-Time Detection of Unauthorized Connections in Low Voltage Power Distribution

Electricity theft in low-voltage distribution networks not only results in financial losses but also degrades power quality, posing significant risks to sensitive consumer electronics and smart home appliances due to voltage fluctuations. Connecting other appliances before the energy meter or short-circuiting the input and output terminals of an energy meter are two prevalent methods of compromising the reliability of the residential energy supply. To detect such tampering, this study presents a consumer-centric Edge ML solution for real-time theft detection that sends a small test signal through the customer’s power line. By adding the traps, this injected test signal is limited between the electric pole and the energy meter’s output terminal. The detection of power theft is greatly aided by this injected test signal. The consumer’s line current is measured using a current transformer (CT). The test signal is separated from the other line components using a bandpass filter, and its features are extracted using the fast Fourier transform (FFT), standard deviation (STD), and kurtosis. To enable low-latency inference at the consumer edge, the trained model is optimized and deployed on a microcontroller-based edge device. The proposed system demonstrated effective real-time detection capabilities and communicated power-theft alerts to service providers via an Internet of Things (IoT) interface. Both offline and real-time evaluations validate the system’s effectiveness in securing the smart residential energy environment.