Title:Online detection of a cascade of multiple change-points

Abstract:We propose novel online detection of a cascading failure in the network using sequential measurement data. Our goal is to detect the failure as quickly as possible after it occurs. To achieve this goal, we propose a temporal diffusion network model representing the dynamics of the potential changes to help us capture the changes as quickly as possible. In cascading failures, once the failure affects a node, its measurements change from a pre-change distribution to an unknown post-change distribution. We develop sequential generalized likelihood ratio statistics that perform joint detection and estimation in detecting the change. Numerical experiments show that our method outperforms existing methods.