A topological model-driven optimization approach for detecting and locating electricity fraud in low-voltage networks

Electricity fraud in low-voltage networks has become increasingly sophisticated, causing both technical and non-technical losses for utilities, along with issues related to power quality and supply continuity. While current artificial intelligence and big data methods are effective for detecting visible fraud, invisible fraud (resulting from direct illegal connections) is harder to detect and locate, often requiring extensive fieldwork. This paper proposes a methodology that leverages network topology, connectivity data, and smart meter measurements to construct a linear model of the low-voltage network and address both visible and invisible fraud. The optimization problem minimizes the differences between recorded and calculated voltages, as well as the discrepancies in power, to detect and locate fraudulent segments. This approach significantly reduces the time required for fraud detection, optimizing efforts and reducing costs for distribution operators. Real-world low-voltage network data tests validate the proposal.