A comparative study of non-parametric models for identification of linear-time invariant systems

Abstract

A comparative study of methods used for identification of Linear-Time-Invariant (LTI) systems based on Non-parametric model (step, impulse, sine and frequency responses) and Parametric model (autoregressive external inputs) was conducted. A series of experiments conducted on a laboratory heating system showed that the step and impulse responses were good identification methods but only limited to first order process. The frequency response using the sine response was good for estimating the gain and phase shift of system in frequency doma-in, but was time consuming. However, the frequency response method using random binary signals was good for unpredicted white noise characteristics and considered the best method for non-parametric system identifica-tion. The autoregressive external input (ARX) model was very useful for system identification, but on applicati-on, few input parameters gave poor model fit, while many parameters led to complex modeling. The analyses were based on Matlab software package Release 14.

Journal of Applied Science and Technology Vol. 13 (1 & 2) 2008: pp. 20-26