Nonlinear System Identification Using Probabilistic Universal Learning Networks

Abstract

Probabilistic Universal Learning Networks are proposed, where a calculation method of the propagation of stochastic signals through Universal Learning Networks is provided. Probabilistic Universal Learning Networks also provide a gradient learning method to optimize parameters in Universal Learning Networks by minimizing the value of the stochastic-based evaluation function. From simulations, it has been shown that identification of a nonlinear dynamic system can be realized without overfitting by using Probabilistic Universal Learning Networks.