Introduction
In science and engineering, system identification is important and has been used widely especially in control system engineering application for inferring models from observed data. It is necessary because these models are able to simulate and predict the actual system behaviour. The study of system identification is necessary since more and more engineering works rely on mathematical models and simulation (Ljung and Glad, 1994; Bequette, 1998).
In system identification, there are few factors need to be considered before designing the experimental procedures for data collection. The choice of the input signal to be used, test signals and sampling rate, must be taken into consideration in order to get the maximum information from the system response. Most researchers in system identification focus or concentrate on two main areas: the selection of model structure that consists of an adequate number of terms in the final model and the estimation of the parameter of those terms. Several researchers reviewed and summarised these problems (Desrochers and Mohseni, 1984; Billings, 1985; Haber and Unbehauen, 1990). Model validation, on the other hand, is also important in the study of system identification in order to verify that the identified models adequately fulfil the modelling requirement according to the objective of model approximation criteria. The basic procedure of system identification can be best described as follows (Ljung, 1999).
- Design an experimental procedure to obtain the process input and output data sets.
- Examine the measured data including removing outliers. Apply filtering to remove measurement and process noise.
- Obtain a set of candidate models based on the experimental data sets.
- Determine the best model from a set of candidate model in step (3) and estimate the model parameter values.
- Validate and evaluate how good the model is. If the model is not satisfactory, repeat step (iv) until all the candidate models have been evaluated.