Researchers build AI system to catch electricity theft in power grids at scale

Computer scientists combined machine learning, deep learning, and graph analysis to detect when people steal power or tamper with meters in smart grids. The system identifies theft patterns across entire grids by watching consumption over time and spotting which houses match known theft signatures — and then flagging similar patterns in connected neighborhoods.

Electricity theft costs utilities billions annually and destabilizes grids, but detecting it at scale has been hard. Most detection systems work on individual meters in isolation. This one watches the spatial patterns — the fact that theft tends to cluster in neighborhoods — which lets it find theft that looks normal on its own but suspicious in context. In practice, this means utilities can move from checking meters one at a time to scanning entire regions for correlated anomalies, catching distributed theft networks instead of individuals.