27 mai 2026 à 13h30
Abstract:
This paper proposes an observer design approach for a class of nonlinear discrete-time systems that applies artificial neural networks. These neural networks are used to calculate the output injection gain and the corresponding Lyapunov function, guaranteeing the stability of the estimation error. A canonical form of Lyapunov function for Lipschitz systems is used to be represented by neural networks. We introduce a locally defined norm-like Lyapunov function when using neural networks to avoid singularity at the origin. Examples of mechanical systems with power nonlinearity illustrate the method’s efficiency.
Inria bulding A, room A00
français
English