A Comparative Study on System Identification Techniques for NARX Model

The relationship between input and output of many nonlinear dynamic systems can be modeled as the Nonlinear Auto-Regressive model with exogenous inputs (NARX model). Identification of the NARX model plays a vital role in system modelling based on NARX models. This paper compares three system identification approaches for the NARX model, including the support vector machine (SVM), the least-square support vector machine and the forward regression orthogonal estimator. Statistical criteria of mean-square error for the one-step-ahead prediction and the model predicted output are used to evaluate the performance of those identification approaches. Simulations of a six degrees of freedom nonlinear system under different levels of noise excitation are given. It seems advisable to choose the SVM for identifying the NARX model.