International Conference «Mathematical and Informational Technologies, MIT-2011»
(IX Conference «Computational and Informational Technologies for Science,
Engineering and Education»)

Vrnjacka Banja, Serbia, August, 27–31, 2011

Budva, Montenegro, August, 31 – September, 5, 2011

Stevovic S.  

Fuzzy logic, neural networks, expert systems and its applicative correlation models

Fuzzy logic, neural networks and expert systems gives an ideal mathematical framework for complex phenomena of optimization in systems engineering. This paper is presenting the mathematical models, which are developed for correlation between technical and non technical system, theoretically and through the case study. Techno economical optimization models are often incomplete, or gives incorrect conclusion, due to lack of all relevant input variables, which have important effect on the result and decision making process, but are not easy to be quantified, such as environmental or social impact of technical solu tion. All non-technical criteria have to be incorporated in the decision-making process as well, from the very first planning step, simultaneously and equally with other technical criteria. Delphi method is one of the possibilities for quantification of non technical input variables. Neural networks are non-linear statistical data modeling tools, used to model complex relationships between inputs and outputs. The fuzzy model for technical system evaluation, proposed in this paper, operates with five input variables. The expert system is trained on 11 different technical subsystem. The input variables, together with corresponding membership functions, are chosen to describe both techno-economic and environmental parameters together with historical and political ones, because those are important for decision making process. Based on the numerical values of technical and quantified non technical input variables, the fuzzification process calculates the values of membership func tion. These membership functions values represent the inputs for the inference decision engine. Based on the set of ten ‘if-then-else’ fuzzy rules, the inference decision machine generates the values of membership functions adjoining the output variable called the grade of the alternative solution. Calibration of the fuzzy model involves tuning the adopted parameters that define the membership functions of input variables and the position of singletons, adjoined to the output variable and definition of fuzzy rules set. The results obtained clearly demonstrate the importance of considering the additional, non-technical criteria in the decision making process by fuzzy expert system model.

To reports list

© 1996-2019, Institute of computational technologies of SB RAS, Novosibirsk