Model-based recurrent neural network for modeling nonlinear dynamic systems

Abstract

A model-based recurrent neural network (MBRNN) is introduced for modeling nonlinear dynamic systems. The topology of MBRNN as well as its initial weights are defined according to the linearized state-space model of the plant. As such, the MBRNN has the ability to incorporate the analytical knowledge of the plant in its formulation. With its original topology intact, the MBRNN can then be trained to represent the plant nonlinearities through modifying its node activation functions, which consist of contours of Gaussian radial basis functions (RBFs). Training involves adjusting the weights of the RBFs so as to modify the contours representing the activation functions. The performance of the MBRNN is demonstrated via several examples. The results indicate that it requires much shorter training than needed by ordinary recurrent networks. This training efficiency is attributed to the MBRNN's fixed topology, which is independent of training.