Robustification of Learning Observers to Uncertainty Identification via Time-Varying Learning Intensity

Abstract

This brief studies the simultaneous estimation of states and uncertainties in general continuous-time systems. In particular, we present a novel time-varying learning intensity (TLI) learning observer (LO). It has the advantage of inheriting the valuable properties of conventional LOs with a simple structure, i.e., the uncertainty estimation is achieved using simply one algebraic equation with low computational costs. The foremost difference in comparison with conventional LOs is the utilization of the TLI approach, which attenuates the overshooting response in the case of large estimation errors and obtains decent performance improvement. Simulations for constant and time-varying signals demonstrate a notable performance boost of TLI-LO.