Candidate
Dongchang Li | University of Waterloo
Title
Quickest Change Detection
Abstract
Quickest change detection (QCD) is a vital procedure of system monitoring that involves optimizing the tradeoff between a measure of detection delay and a measure of the frequency of false alarms. From a filtering point of view, we propose an effective method of tracking the generalized cumulative sum (CUSUM) statistics in practice to detect a change in the hidden Markov models with nonlinear dynamics.
Specifically, we focus on the cases where the change excites deterministic dynamical systems from a pre-change steady state to a post-change steady state, and to post-change periodic oscillations. The performance of the generalized CUSUMs using particle filters, in particular, is examined with Monte Carlo simulations in the pure surge mode of Moore-Greitzer compressor models with an additive change.