V.A. Galkin – Ph.D.(Eng.), Associate Professor, Department «Information Processing and Control Systems», Bauman Moscow State Technical University
S.N. Krasilnikov – Master, Department «Information Processing and Control Systems», Bauman Moscow State Technical University
V.B. Popenkov – Master, Department «Information Processing and Control Systems», Bauman Moscow State Technical University
J.C. Gonsales Gusev – Post-graduate Student, Department «Information Processing and Control Systems», Bauman Moscow State Technical University
When designing a control system, it is possible to use fuzzy inference systems because of the possibility of operating with fuzzy input da-ta and working with a degree of data reliability. Systems with fuzzy logic convert the values of the input variables of the process in ques-tion into output variables based on the use of fuzzy production rules. Such systems include a base of rules for fuzzy products and imple-ment a fuzzy conclusion.
The fuzzy inference process is divided into 5 stages: Formation of the rule base of fuzzy inference systems; Fuzzing input variables; Ag-gregation of sub-conditions in fuzzy product rules; Activation or composition of sub-conclusions in fuzzy product rules; Accumulation of conclusions fuzzy rules of products.
Varying the parameters on each of them determines a certain algorithm. There are various fuzzy inference algorithms.
In order to choose the most suitable algorithm for a specific task, you need to take into account its specificity. To solve the problem of finding a real assessment of the quality of the «Internet access» service, it is necessary to use such parameters as: level of use, speed, delay, network errors, time window of previous quality estimates, grouping users based on their subjective requirements for the service.
After a comparative analysis of two fuzzy inference algorithms (the Mamdani algorithm and the Sugeno algorithm), we can conclude that they have differences in the knowledge base format and the defasification procedure. For problems where the most important is the ac-curacy of identification of nonlinear dependencies, the most suitable is Sugeno's algorithm.
Therefore, to solve the problem of finding a real assessment of the quality of the «Internet access» service, where possible 11! varia-tions of parameters, Sugeno’s algorithm should be used, since for large volumes of sampling of experimental data, identification using Sugeno’s model provides greater accuracy.
To confirm this choice, the ANFIS learning process was modeled to assess QoE in the Matlab system. Networks of the ANFIS type were created based on the Mamdani and Sugeno algorithms. After their training, the accuracy of the algorithms was compared. As a result, it was found that Sugeno’s algorithm should be used.
- Yarushkina N.G. Osnovy teorii nechetkikh i gibridnykh sistem: Ucheb. posobie. M.: Finansy i statistika. 2004. 320 s.
- Kruglov V.V., Borisov V.V. Gibridnye neironnye seti. Smolensk: Rusich. 2001. 224 s.
- Kruglov V.V. Sravnenie algoritmov Mamdani i Sugeno v zadache approksimatsii funktsii. 2001. http://sgma.alpha-design.ru/MMORPH/N 7- html/KRUGLOV/kruglov.htm.
- Sistemy nechetkogo vyvoda. URL = http://fuzzy-group.narod.ru/files/Fuzzy_Modeling/Lection08.1.Fuzzy.inference.system.pdf.
- Shtovba S.D. Proektirovanie nechetkikh sistem sredstvami MATLAB. M.: Goryachaya liniya – Telekom. 2007. 288 s.
- Leonenkov A.V. Nechetkoe modelirovanie v srede MATLAB i fuzzyTECH. SPb.: BKhV-Peterburg. 2003. 736 s.
- Kuzkin A.A. Otsenivanie pokazateleй̆ effektivnosti i rezultativnosti IT-protsessov s ispolzovaniem gibridnykh neiro-nechet¬kikh setei. Naukovedenie (Internet-zhurnal). 2014 (yanvar–fevral). № 1. URL = http://naukovedenie.ru/PDF/57TVN114.pdf.