ISSUE 181, article 5

DOI:https://doi.org/10.15407/kvt181.01.056

Kibern. vyčisl. teh., 2015, Issue 181, pp.

Melnichuk S.V.

Space Research Institute National Academy of Sciences of Ukraine and State Space Agency of Ukraine

METHOD OF STRUCTURAL PARAMETRIC MULTIVARIABLE SYSTEM IDENTIFICATION USING FREQUENCY CHARACTERISTICS

Introduction. One of the important directions in the identification of linear systems are frequency domain methods. In recent decades a finite-frequency approach, focused on the use under bounded uncertainty has been developed. Within finite-frequency approach a method, that allows to construct models with reduced dimensionality has been proposed. The method includes a step of structural identification with regularization by model dimension. This method was used to identify single-input single-output (SISO) systems, so it could not be applied to systems with multiple input and multiple output (MIMO).
Purpose. In order to generalize the method it is proposed to identify SISO models of subsystems, that describes individual inputs and outputs, and then combine them. The main purpose of research is to develop an algorithm, that combine separate SISO models into one general MIMO model.
Results. Separate SISO models determined by their invariant properties. As simple combination of SISO models leads to a MIMO model of large dimension, and some invariant properties in different models may be similar, it makes sense to carry out unification by equating this invariants.
Possibility of association for different combinations of SISO models, that have the same eigenvalues were investigated. It is shown that by combining models additional dependencies between coefficients may be imposed. It is shown that if the dependency graph contains no cycles, then the union is possible. On the basis of this fact the synthesizing algorithm was proposed.
Conclusions. The proposed identification algorithm builds the general MIMO model from separate SISO models so that the dimension of resulting model may be significantly less, than sum of dimensions of original SISO models. The proposed algorithm saves all invariant characteristics of the original models, so approximation accuracy by the each input-output relation is stored.

Keywords: System identification, frequency domain, structural identification, reduced dimensionality.

Download full text (ru)!

References

1 Orlov Y.F. Frequency Parameter-Based Identification at Parallel Testing. Automation and Remote Control, 2007, vol.68, no. 1. pp. 18–37.

2 Orlov Y.F. Frequency Parameter-Based Identification at Parallel Testing. Avtomatika i Telemekhanika, 2007, vol.68, no. 1. pp. 20–40 (in Russian).

3 Alexandrov A.G. Method of Frequency Parameters. Avtomatika i Telemekhanika, 1989, vol 50, no. 12. pp. 3–15 (in Russian).

4 Alexandrov A.G., Orlov Y.F. Finite-frequency identification of multidimensional objects. 2-nd Russian-Swedish Control Conference. St. Petersburg, 1995, pp. 65–69 (in Russian).

5 Gubarev V.F., Melnychuk S.V. Identification of Multivariable Systems Using Steady-State Parameters. Journal of Automation and Information Sciences, 2012, vol. 44. i. 9. pp. 24–42.

6 Gubarev V.F., Melnychuk S.V. Identification of Multivariable Systems Using Steady-State Parameters. Journal of Automation and Information Sciences, 2012, no 5. pp. 26–42 https://doi.org/10.1615/JAutomatInfScien.v44.i9.30 (in Russian).

7 Melnychuk S.V. Regularity Investigation For Multidimensional System Identification Problem by the Frequency Method. Cybernetics and Computer Engineering, 2014, no 176. pp. 19–33 (in Russian).

8 Melnychuk S.V. Modified Frequency Method of Structural-Parametric System Identification. Journal of Automation and Information Sciences, 2015, no 4. pp. 27–36 (in Russian). https://doi.org/10.1615/JAutomatInfScien.v47.i8.60

Received 05.06.2015