Core Concepts
The authors propose a moving horizon estimation (MHE) scheme to robustly estimate the states and time-varying parameters of nonlinear systems, where the observability of the parameters may depend on the system's excitation and can be absent during operation.
Abstract
The authors consider a nonlinear discrete-time system with states x and time-varying parameters z. They assume that the states x are uniformly detectable (i-IOSS) while the parameters z are non-uniformly observable, meaning their observability depends on the system's excitation and may be absent during operation.
To address this challenge, the authors propose an MHE scheme that involves a standard quadratic cost function with an adaptive regularization term. This regularization term depends on the current observability of the parameters, which is monitored online. The authors develop robustness guarantees for the overall estimation error that are valid for all times and improve the more often the parameters are detected to be observable during operation.
The key aspects of the proposed approach are:
It does not require a priori guarantees on the observability of the parameters, which is usually impossible for general nonlinear systems.
It relies on online monitoring of parameter observability and selects an appropriate regularization term accordingly.
It provides robustness guarantees for the state and parameter estimation errors that are valid independent of the parameter observability and improve with more frequent parameter observability.
It is illustrated through a simulation example of a modified Chua's circuit system with time-varying parameters.