A Tunable Fraud Detection System for Advanced Metering Infrastructure Using Short-Lived Patterns

We propose a fraud detection systems (FDS) for advanced metering infrastructure based on anomaly detection on the energy consumption reports from smart meters. Fraud inspection is triggered when a discrepancy between the energy supplied by the grid and that reported by smart meters is detected. We use an innovative approach, where consumption reports registered shortly before and after the discrepancy detection are compared to detect a fraud. Our FDS introduces an important innovation by showing that it is possible to use only a small set of recent measures to define a consumption pattern. We call these patterns short-lived because they are expected to represent the behavior of a consumer for a short period, only enough to detect an ongoing fraud. This approach allows the FDS to account for natural changes in the consumption behavior of users and also helps to preserve their privacy. The FDS can be tuned, using an optimization procedure, by imposing constraints on the true or false alarm rates, or maximizing an objective function that represents the revenue of the utility.