Adaptive Estimation for Nonlinear Systems using Reproducing Kernel Hilbert Spaces
Provides a theoretical foundation for adaptive estimation of nonlinear systems using RKHS, benefiting control and estimation communities, but remains theoretical without empirical validation.
This paper extends a framework for online adaptive estimation of unknown nonlinear ODEs by representing the unknown function in an RKHS, deriving stability and convergence conditions for infinite- and finite-dimensional estimators, and introducing a new persistency of excitation condition. The approach is validated for exponential and multiscale kernels.
This paper extends a conventional, general framework for online adaptive estimation problems for systems governed by unknown nonlinear ordinary differential equations. The central feature of the theory introduced in this paper represents the unknown function as a member of a reproducing kernel Hilbert space (RKHS) and defines a distributed parameter system (DPS) that governs state estimates and estimates of the unknown function. This paper 1) derives sufficient conditions for the existence and stability of the infinite dimensional online estimation problem, 2) derives existence and stability of finite dimensional approximations of the infinite dimensional approximations, and 3) determines sufficient conditions for the convergence of finite dimensional approximations to the infinite dimensional online estimates. A new condition for persistency of excitation in a RKHS in terms of its evaluation functionals is introduced in the paper that enables proof of convergence of the finite dimensional approximations of the unknown function in the RKHS. This paper studies two particular choices of the RKHS, those that are generated by exponential functions and those that are generated by multiscale kernels defined from a multiresolution analysis.