Наука и техника (May 2017)

PARAMETRIC IDENTIFICATION OF STOCHASTIC SYSTEM BY NON-GRADIENT RANDOM SEARCHING

  • A. A. Lobaty,
  • V. Y. Stepanov

DOI
https://doi.org/10.21122/2227-1031-2017-16-3-256-261
Journal volume & issue
Vol. 16, no. 3
pp. 256 – 261

Abstract

Read online

At this moment we know a great variety of identification objects, tasks and methods and its significance is constantly increasing in various fields of science and technology. The identification problem is dependent on a priori information about identification object, besides that the existing approaches and methods of identification are determined by the form of mathematical models (deterministic, stochastic, frequency, temporal, spectral etc.). The paper considers a problem for determination of system parameters (identification object) which is assigned by the stochastic mathematical model including random functions of time. It has been shown that while making optimization of the stochastic systems subject to random actions deterministic methods can be applied only for a limited approximate optimization of the system by taking into account average random effects and fixed structure of the system. The paper proposes an algorithm for identification of parameters in a mathematical model of the stochastic system by non-gradient random searching. A specific feature of the algorithm is its applicability practically to mathematic models of any type because the applied algorithm does not depend on linearization and differentiability of functions included in the mathematical model of the system. The proposed algorithm ensures searching of an extremum for the specified quality criteria in terms of external uncertainties and limitations while using random searching of parameters for a mathematical model of the system. The paper presents results of the investigations on operational capability of the considered identification method while using mathematical simulation of hypothetical control system with a priori unknown parameter values of the mathematical model. The presented results of the mathematical simulation obviously demonstrate the operational capability of the proposed identification method.

Keywords