Zhukov A. Identification of the different nature complex systems in the uncertainty conditions

Українська версія

Thesis for the degree of Candidate of Sciences (CSc)

State registration number

0410U003221

Applicant for

Specialization

  • 01.05.04 - Системний аналіз і теорія оптимальних рішень

20-04-2010

Specialized Academic Board

Д 26.002.03

Educational and Scientific Complex "Institute for Applied System Analysis" of National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"r

Essay

The dissertation is devoted to the development and scientific prove of system approach to structure and parametric identification of the different nature complex systems in the prior uncertainty conditions and based on it the Recursive Identification Method (RIM), approaches and algorithms of approximate models iterative recovering that are suitable for applied system analysis and optimal decisions theory problems solving are developed. The structure of the dissertation is as follows: introduction, five chapters, conclusion, references and 2 appendixes. In the introduction to the dissertation the author stresses out the reasons of choosing the theme of dissertation, innovations produced, the approbation of the researcher's results (conferences, publications etc.). In the first chapter the problem of complex systems identification in uncertainty conditions is analyzed and its relation to the experimental data interpretation is pointed out. The main research direction in dissertation is focused on development of the system approach to structure and parametric identification of the complex systems in the bounded uncertainty case. In the second chapter the structural singularities of multivariable linear dynamical systems and their affect to the quality of recovered approximation models are explored; the hierarchical structure and components of the system approach to identification of the complicated multivariable different nature systems at the presence of uncertainty is developed. The approach is based on the analyzing of the systems identification task and corresponding identification methods; giving of approaches to the output data uncertainty formalization; researching of the difficulties that take place in systems identification in bounded uncertainty case; developing of the new method (RIM) and corresponding algorithms that have formed the core of applied implantation. The basic idea of RIM was proposed by V.F. Gubarev and is in consequent recovering of approximation model via separate parts - submodels that contain clusters of system modes, i.e. generalized degrees of freedom. The rationality of the recovered model structure in the state space with the Jordan canonic form is proved. The offered approach to model structure selection allows to achieve the independence between the subsystems-submodels and to transform the internal relationships to external ones. This leads to simplifying of the submodels aggregation procedure. The rectangular impulses using in the offered RIM-method as the input exciting influence, in spite of the widely used in 4SID-methods persistently exciting white noises, is recommended. The analytical expression of input-output correlations in matrix form for given type of input influence that are obtained has formed the core of the offered method RIM. In the third chapter the recursive identification method and algorithms for systems with scalar input and output, based on system approach, is developed. The system response modal expansion that allows to realize the developed RIM, is obtained. The informatyability of system modes is defined and approach to informative data selection that provides the minimal dimension of recovered approximation submodels and identification procedure stability, is introduced. The sequence of loop-iterated steps that constitutes the core of developed RIM, is offered. The resulting approximation model is recovered by aggregation of identified submodels. The small dimension of recovered submodels allows to overcome the incorrectness problem caused by the large system dimension, and to achieve the simplest way of model order definition. The two steps algorithm of submodel structure and parameters identification is developed. The difference filters with tuning parameters that allows to define the submodel spectrum are offered. The results of the offered RIM experimental research, that confirm its usability, are illustrated. The regularization property of the developed method RIM, i.e. recovered approximation model is stable and comparable with the measure of presence uncertainty in initial data, is proved. The comparative analysis of the developed RIM-method with traditional 4SID-method is presented. It is discovered that in bounded uncertainty identification in some cases the informative data selective approach used in the RIM-method allows to recover approximation models of higher order that contain except slow also some fast attenuated modes, unlike the 4SID-method. At the same time, in identification of the systems with slowly attenuated modes, the 4SID-method is also acceptable. In the fourth chapter the generalization of RIM to the multi-input - multi-output system identification case is considered. The special averaging algorithms to solve the task of observation and control system matrices estimation using obtained modal coefficients values are developed. In the fifth chapter the application of research results obtained in dissertation are presented. The developed system approach, recursive identification method and algorithms have used by NASU and NSAU Space Research Institute in the autonomous underwater robot navigation and the space-craft orientation control problems solving. The important applied problem of the space-craft inertial characteristics estimation is solved.

Files

Similar theses