Continuous-time nonlinear system identification using neural network

Abstract

In this paper, a continuous-time neural network nonlinear system identification algorithm using the system input/output signals is developed for a class of nonlinear systems. In control applications, the continuous-time nonlinear system model is more truthful for the original nonlinear process compared to the widely used discrete-time neural network model. In the identification algorithm, a canonical form is selected to represent the identified system. The identification algorithm consists of two stages: (i) preprocessing the system input and output data to estimate the state variables in the chosen model coordinate; (ii) neural network parameter estimation. Discrete-time implementation of the developed algorithm is introduced. Identification examples are illustrated with a single-input-single-output benchmark model and a hardware-in-loop multi-input-multi-output 3 degrees-of- freedom differential thrust flight control testbed.