Comparison of different methods for identification of industrial processes

  • Authors:

  • I. Gustavsson

  • Affiliations:

  • Division of Automatic Control, Lund Institute of Technology, Lund, Sweden

  • Venue:
  • Automatica (Journal of IFAC)
  • Year:
  • 1972

Quantified Score

Hi-index 22.16

Visualization

Abstract

Plants have been modelled using different identification methods. Some results from the identification of the dynamics of a nuclear reactor, a distillation column, a superheater and a paper machine are presented. Models for the different processes, obtained by the maximum likelihood method, are compared to models, obtained by other identification methods: the least squares method, the tally principle method, correlation analysis, and spectral analysis. Comparisons are made between the model parameters, and between the transient and frequency responses of the different models. It is pointed out that such comparisons may not be relevant for the performance of control strategies, synthesized from the models. However, the comparisons give insight into the properties of different identification methods. The problem of choosing model order for parametric identification methods is discussed. Simulations have also been used in order to compare the performance of the methods on data from known processes. The maximum likelihood method turns out to be superior, especially for low signal to noise ratios.