Oliinyk A. Methods of diagnostic model synthesis based on computational intelligence

The thesis is devoted to the solution of the theoretical and practical problem of the development and research of methods of diagnostic model synthesis, which combine principles of intelligent computations and parallel computing, allowing to accelerate the process of construction of diagnostic models, to raise its interpretability and generalization abilities. The object of research is the process of data-based diagnostics. The subject of research is the methods of model synthesis in intelligent diagnostic systems. The process of diagnostic model synthesis on the basis of neuro-fuzzy networks is analyzed and researched. In the thesis the stochastic method of model synthesis, based on decision trees, was proposed. It uses information about feature informativeness, complexity of synthesized tree and also about recognition accuracy. The stochastic method of extraction of numerical association rules and method of synthesis of production rules based on negative selection in the case of non-uniform distribution of instances of sampling classes were developed. The parallel method of extraction of production rules based on computational intelligence was developed. The feature informativeness estimation criteria and the parallel stochastic method of data reduction were developed. The model of the process of parametric synthesis of neuro-fuzzy networks in a tiered-parallel form was proposed. The method of parametric identification of neuro-fuzzy networks based on parallel random search was developed. The method of additional training of diagnostic neuro-fuzzy models was proposed. The experimental research of the developed methods by models synthesis for solving real-world problems of diagnostics was carried out. The practical use of the dissertation results is confirmed by the acts of implementation, which prove the correctness of the theoretical positions of the dissertation and the high efficiency of the developed methods.