A new method for parameter estimation in the NARMAX model using neural computation

Abstract

This work shows a new method for parameter estimation in the NARMAX (Non-linear AutoRegressive Moving Average with eXogenous inputs) model using neural computation. A three-layered feedforward neural network is trained to describe a system. The actual input of the system and the computed output of the network are used as the input data set of the network for training. Parameters in the NARMAX model are calculated from the values of weights and the sigmoid functions in neural units expanded in a series by Maclaurin's formula. The structure of the NARMAX model is finally determined by the Baysian information criteria. The proposed method, therefore, requires no prior knowledge of the structure of the NARMAX model.