Recent Advance on State Estimation of Stochastic Hybrid Systems

Abstract

This presentation summarizes some recent progress on observability and observer design for stochastic hybrid systems in which all subsystems are unobservable. Such hybrid systems capture the emerging technologies on networked systems in which capabilities of individual sensing devices are highly limited and cannot provide sufficient information for estimating the entire states of the system. A central operator needs to combine information from different sensing systems to obtain information on the states of the entire system. The notion of stochastic observability, its probabilistic descriptions, design methods for subsystem observers, and their organization for estimating the entire state are discussed. Convergence properties are established, including strong convergence and exponential convergence rate. Estimation error probabilities under finite data are derived by using the large deviation principles.