Quickest Detection of Moving Anomalies in Sensor Networks

The problem of sequentially detecting a moving anomaly is studied, in which the affects different parts sensor network over time. Each characterized by pre- and post-change distribution. Initially, observations each are generated according to corresponding pre-change After some unknown but deterministic time instant, emerges, affecting sets sensors as progresses. Our goal design stopping procedure detect emergence quickly possible, subject false alarms constraints. studied quickest change detection framework where it assumed that evolution deterministic. A modification Lorden's delay proposed account for trajectory maximizes procedure. It established Cumulative Sum-type test solves resulting sequential exactly when homogeneous. For case heterogeneous sensors, scheme can be modified provide first-order asymptotically optimal algorithm.