Comparison of state estimation techniques for nonlinear hybrid systems

Abstract

Modern real-world engineering systems exhibit complex hybrid behaviors, which are composed of continuous nonlinear plant dynamics interspersed with discrete mode switching. State estimation of hybrid systems for accurate and timely online monitoring and diagnosis applications is a difficult task. A number of different methods have been proposed, and we develop a conceptual framework to perform a comparative study of four different hybrid state estimation algorithms in this paper: (1) the switched extended Kalman filter; (2) focused hybrid estimation; (3) multiple-modal particle filtering; and (4) the one-step look-ahead particle filtering. The conceptual comparison is followed by an empirical evaluation, where we study the effectiveness of these algorithms in tracking the behaviors of a Reverse Osmosis Subsystem of an Advanced Water Recovery System that was developed at the NASA Johnson Space Center. We discuss our results, which show the strengths and weaknesses of each of these algorithms, and propose topics for further research into this important problem.