Nonlinear system identification based on NARX network

This paper discusses identification of nonlinear system with nonlinear AutoRegressive models with eXogenous inputs (NARX). NARX network is a dynamic neural network which appears effective in the input-output identification of both linear and nonlinear systems. When identifying them by NARX model, the first step is to collect training data and the final results vary considerably with different training data. The paper compares the training results of three kinds of signals, including SPHS signal, Gaussian white noise and mixed signal. Our results show the response characteristics of NARX model trained by different signals can be used to design the input training signal.