Detection and localization of energy theft in distribution networks using artificial intelligence neural networks

This study investigates the pervasive issue of non-technical losses, specifically energy fraud, and theft, in distribution networks. It proposes a new approach for the accurate detection and localization of theft incidents using an artificial intelligence neuron network and the forward propagation method. A simulated smart network is utilized, comprising smart energy meters and a central server-connected information processing unit. Technical losses are determined and optimal voltage drop values are calculated through analysis of load data, voltage drop measurements, and energy flow data. The proposed neuron network algorithm in MATLAB compares ideal and actual voltage dips to identify theft locations. Comparative analysis with industry-standard equation-based algorithms demonstrates the superior detection accuracy of the neuron network approach. Unlike fixed algorithms based on predetermined voltage drop values, the neuron network exhibits adaptability to changing load conditions, offering a robust and reliable theft detection mechanism. Continuous training of the neuron network using anticipated data is essential for sustaining optimal performance. Regular updates incorporating relevant network data and customer load profiles can enhance theft detection accuracy and efficiency. The findings of this study contribute to the ongoing efforts in combating energy theft and fraud in distribution networks. The proposed neuron network-based methodology provides a promising solution for precise theft localization, empowering energy providers to effectively mitigate non-technical losses. Future research can explore innovative techniques to further improve the neuron network's training process, aiming for enhanced overall performance and higher detection rates. Keywords: Theft Energy, Neural Network, Distribution System, Transformer, theft current, Drop voltage.