R.A. Tomakova - Dr.Sc. (Eng.), Professor, Southwest State University
Mohammed Ali Awad - Post-graduate Student, Southwest State University
L.V. Pleckanos - Ph.D. (Eng.), Associate Professor, Southwest State University

The idea of hierarchical classifier construction realizing principles of multiagent intelligent systems is proposed in this article. The proposed model of hierarchical classifier works on the basis of probabilistic neural networks and fuzzy logic of decisions making. The analysis of possible designs of its separate functional elements has been carried out in this work. The model differs in the fact that hybrid neural network consisting of macrolayers with PNN-FNN structure is used as hierarchical multiagent classifier at two-alternative classification. The first macrolayers has been constructed on the basis of probabilistic neural network and the second - on the basis of fuzzy neural network. The has been constructed structural-functional model of biomaterials electric conduction in abnormal zones whish is based on the hypothesis of reversible dielectric break-down and "biomaterial response reaction" and biomaterial which is fulfilled in the form of serially being adjusted diode-resisting cells. The proposed model allows to forecast biomaterial electric conduction depending on frequency and value of probing current. While simulating BAP volt-ampere characteristic there has been used hybrid approach consisting of two stages. At the first stage functional model aggregating expert knowledge about functioning of the object under investigation is constructed, and at the second stage there is constructed mathematical model coordinating the process of model functioning and experimental data. BAPs processing maximal accessibility were chosen to conduct experimental investigations. There has been developed method of forming space for informative characters meant for classifiers of biomaterial functional state. There has been also shown expediency of biotissue volt-ampere characteristics in each bioactive point of pair meridians by the way of constant stabilized by a level voltage action on them. Coefficients of approximating polynomial for each of volt-ampere characteristics are used as the input vector for neural network. Instruction of neural network is based on diagnostic decisions making according to selected classes of diseases. The process of simulation was carried out in MATLAB 7.13 medium.

 

1. Rassel S., Norvig P. Iskusstvenny'j intellekt: sovremenny'j podxod / per. s angl. M.: Izdatel'skij dom «Vil'yams». 2006. 1408 s.
2. Korenevskij N.A., Filist S.A., Ustinov A.G. i dr. Geometricheskij podxod k sintezu nechetkix reshayushhix pravil dlya resheniya zadach prognozirovaniya i mediczinskoj diagnostiki // Biomediczinskaya radioe'lektronika. 2012. № 4. S.20-26.
3. Filist S.A., Korenevskij N.A., Shatalova O.V. i dr. Diagnosticheskaya sistema na osnove analiza vol'tamperny'x xarakteristik bioaktivny'x tochek // Biotexnosfera. 2013. №5(29). S. 33-38.
4. Klyuchikov I.A., Kabus Derxim ali Kassim, Shatalova O.V., Jaa Zar Do Parametricheskie modeli bioimpedansa dlya identifikaczii funkczional'nogo sostoyaniya zhivoj sistemy' // Biomediczinskaya radioe'lektronika. 2012. № 4. S.50-56.