Dealing with Uncertainty in the Hybrid World

This paper presents an efficient state estimation algorithm for hybrid systems based on a least-squares Interacting Multiple-Model setup. The proposed algorithm is shown to be computationally efficient when compared with the Moving Horizon Estimaton algorithm that is a brute force optimization algorithm for simultaneous discrete mode and continuous state estimation of a hybrid system. The main reason has to do with the fact that the proposed algorithm is able to disregard as many discrete mode sequence estimates as possible. This is done by rapidly computing good estimates, separating the constrained and unconstrained estimates, and using some auxiliary coefficients computed off-line. The success of this state estimation algorithm is shown for a fault detection problem of the benchmark AMIRA DTS200 three-tanks system experimental setup.