Modeling of nonlinear block-oriented systems using orthonormal basis and radial basis functions

This paper presents a methodology for modeling the Wiener, Hammerstein and feedback-nonlinear systems via orthonormal basis and radial basis functions. The approach is computationally effective, in particular, in terms of elimination of the disastrous bilinearity effect due to the use of regular or inverse orthonormal basis functions to model the linear dynamic block. Scaling parameters of orthonormal basis and radial basis functions are updated recursively using the stochastic gradient method. The modeling of a nonlinear static block with radial basis functions is particularly recommended for the Hammerstein and feedback-nonlinear systems. A simulation study for the magnetic levitation process confirms the attractiveness of the approach.