Nedashkovskaya N. Methodology and decision support system on basis of hierarchical and network decision models

Українська версія

Thesis for the degree of Doctor of Science (DSc)

State registration number

0518U002662

Applicant for

Specialization

  • 01.05.04 - Системний аналіз і теорія оптимальних рішень

04-12-2018

Specialized Academic Board

Д 26.002.03

National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”

Essay

In the dissertation work, an important scientific and technical problem has been solved, which deals with development of mathematical and methodological support for increasing the reliability of solutions to decision analysis problems in complex weakly structured systems based on hierarchical and network models. The scientific novelty of the work is determined by the following theoretical and practical results obtained by author. Using proposed systematic approach, a new methodology of decision support is developed, which allows to increase the reliability of solutions of decision analysis problems in complex weakly structured systems on the basis of hierarchical and network models. This methodology includes the proposed and described below methods and techniques. A new method for evaluating and improving the consistency of expert judgements, which are given in a form of pairwise comparison matrix, is developed. Features of the method include an analysis of property of weak inconsistency, the presence of cycles in a pairwise comparison matrix and a search for the most inconsistent element of this matrix. The method can be applied to pairwise comparison matrices of various types, including multiplicative, additive, fuzzy and other. A Transitiv method for searching the most inconsistent elements of the matrix is proposed. A method of flows for finding the most inconsistent element of the matrix is improved by taking into account the input flow. The simulation shows that the developed Transitiv method and the method of flows are more efficient than existing methods. Usage of the proposed method of consistency evaluating and improving allows to obtain pairwise comparison matrices of acceptable quality for all elements of the model and these matrices can be used further to find local weights of model’s elements. A new method for calculating confidence intervals of local weights is developed, which, unlike others, takes into account the uncertainty of scale, expert's personal qualities such as optimism and pessimism, and does not require comparison of groups of elements with the frame. The method is based on notions of the Dempster-Schafer theory of evidence and results of computer simulation of expert's judgments. An uncertainty index of expert judgments is proposed, assuming that this uncertainty is caused by above factors. An improved method for calculating fuzzy local weights on basis of fuzzy pairwise comparison matrix is proposed, which differs from others in estimating and increasing the consistency of the matrix and taking into account properties of weak and strong order pre-servation on a set of calculated fuzzy weights. This method, unlike existing ones, makes it possible to determine the weak inconsistency of fuzzy matrix, to assess the acceptability of inconsistency level of fuzzy matrix for reliable local weights calculation, and to find the most inconsistent elements of the matrix using methods developed for crisp matrices. A hybrid method for calculating aggregated weights of hierarchical model elements with interdependent decision criteria has been improved, when input data for evaluation are fuzzy expert judgments. Improvement consists in using the developed more effective methods for assessment and increasing of crisp and fuzzy expert judgements consistency. A method for complex sensitivity analysis of results has been improved by taking into account sensitivity analysis of local rankings of model’s elements. In the developed method for estimating local sensitivity, intervals and indices of stability of pairwise comparison matrix elements are calculated, which retain the best decision alternative and all ranking of alternatives. Resulting stability intervals allow to find critical elements of the problem that require more careful analysis. A new technique for estimating the rank reversal is suggested, which can appear after applying combination rules of confidence functions for model’s elements. Using this tech-nique, the Dempster, Yager, Zhang, Dubois and Prada and other combination rules were examined. Cases and features of rank reversals appearance in these rules were revealed. New techniques and tools for modeling a process of decision alternatives evaluation by an expert of high competence, expert-optimist and expert-pessimist while performing pairwise comparisons are developed. Using these techniques and tools, efficiency of proposed methods has been proved. A decision support system has been constructed on basis of proposed methods and techniques. This system has been used to solve several practical problems.

Files

Similar theses