Guda A. Ensemble search models and methods for parametric identification of systems with chaotic behavior

Українська версія

Thesis for the degree of Doctor of Science (DSc)

State registration number

0518U000113

Applicant for

Specialization

  • 01.05.02 - Математичне моделювання та обчислювальні методи

12-04-2018

Specialized Academic Board

Д 08.084.01

National Metallurgical Academy Of Ukraine

Essay

In order to improve the identification system speed, a structure. which consist of a variety search agents and a search coordinatoris is proposed. Methods for estimating the parameter of the triplet of search agents are created, both with the direct use of the criterion and with the use of the quality function. The dependence of the identification error on the method and its parameters is investigated. A group of methods for the work of the search coordinators has been created, the dependencies of the identification quality on the parameters of the coordinator's interaction with the agents have been determined. A classification of system identification systems has been created, which allows not only briefly describe the set of methods used, but also allows us to create new identification systems when new methods of agents and coordinators appear. The properties of the set of proposed methods for an artificial quasistationary test system are investigated. A subset of methods is selected, which it is expedient to use in further research. The "qontrol" software was created, which is intended for modeling complex nonlinear dynamical systems, including in the analysis of chaotic dynamics systems processes identification. The working capacity and properties of the developed methods have been verified in solving the problem of identifying several nonlinear systems, both known chaotic dynamics systems, and new ones. It was determined that there are systems that are not chaotic, but from the point of view of identification they have common features. The influence of the parameters of the identification system on the quality of its work is investigated, which gives grounds for the adaptation of these parameters. It is determined that the type of criterion and the dynamics of agents have the greatest influence. The properties of the proposed identification methods are investigated on two real physical objects: the Kolpitts oscillator and the system of coupled relaxation elements. The methods have confirmed their efficiency, and it is found that the dependence of the quality of identification on its parameters for real systems has the same form as for the test ones.

Files

Similar theses