"Distributed state estimation with moving horizon
observers."
Marcello Farina, Politecnico di Milano, Italy
Abstract
Nowadays, the application of state estimation
and control algorithms to large-scale, multi-agent and
distributed systems is an issue of paramount importance.
Practical problems in processing and transmitting large
amounts of data limit the applicability of optimal
centralized methods. Besides, robustness, fault-tolerance
and reconfigurability requirements promote the study of
distributed algorithms. This is also spurred by the
availability of low cost sensing devices, endowed with
computational capabilities, which can coordinate their
activity through wireless communication networks (i.e.,
sensor networks).
Among the open problems, we address the use of sensor
networks for distributed state estimation. The main
challenge is to provide a methodology which guarantees
that all the sensors asymptotically reach a common
reliable estimate of the state variables.
A distributed algorithm based on the concept of Moving
Horizon Estimation (MHE) is proposed. This approach has
many advantages; first of all, the observer is optimal in
a sense, since a suitable minimization problem must be
solved on-line at each time instant. Furthermore, we prove
that, under weak observability conditions, convergence of
the state estimate is guaranteed in a deterministic
framework. Finally, constraints on the noise and on the
state are taken into account, as it is common in receding
horizon approaches in control and estimation.
Slides
