Probabilistic state estimation for labeled continuous time Markov models with applications to attack detection

Abstract

This paper is about state estimation in a timed probabilistic setting. The main contribution is a general procedure to design an observer for computing the probabilities of the states for labeled continuous time Markov models as functions of time, based on a sequence of observations and their associated time stamps that have been collected thus far. Two notions of state consistency with respect to such a timed observation sequence are introduced and related necessary and sufficient conditions are derived. The method is then applied to the detection of cyber-attacks. The plant and the possible attacks are described in terms of a labeled continuous time Markov model that includes both observable and unobservable events, and where each attack corresponds to a particular subset of states. Consequently, attack detection is reformulated as a state estimation problem.