THE ARCHITECTURE OF THE HYBRID EXPERT SYSTEM
Gavrilov A.V., Novitskaya Y.V.
Novosibirsk State Technical University, Novosibirsk, firstname.lastname@example.org
In the last years there has been a growing interest in the use of artificial intelligence in a wide range of various related applications. The conventional artificial intelligence facilities are the expert systems and the artificial neural network possessing determines advantages and disadvantages. The intelligent hybrid systems allow to use advantages of mentioned artificial intelligent methods, to overcomes some of the major disadvantages of its components and can provide solutions to problems that are not solvable by an individual intelligent technique alone [1, 2].
Perhaps, the most important class of intelligent hybrid systems is the hybrid expert systems - integration of expert systems and neural networks. The hybrid expert system allowing to capture both of types of information, with the formal knowledge being contained in the expert system and the informal knowledge being contained within the neural network. Thereby the hybrid expert systems efficient in medicine, finances, in the pattern recognition and etc [3, 4].
The hybrid expert system architecture proposed in this paper consists of fuzzy expert system and artificial neural network. Fuzzy assigned and processed with using the certainly factors and linguistic variables. The linguistic variables offered by fuzzy representations allow pseudo-verbal description close to natural human expression .
The knowledge base consists of containing rules and frames-classes constant part and containing frames-instances database.
Frames allow describe application domain in hierarchies of classes and owners. Frames consist of slots, which can be symbol, numerical, linguistic variable, date and time. Frame can be joined with rules and procedures, processing determined events. Also in the backward inference process can be used data received from the preliminary training neural network. Neural network allows extracting facts to the expert systems database (knowledge acquisition) from data and knowledge sources.
To integrate these different facilities into one system it is necessary that the data representations are uniform in all subsystems. Receiving from the neural networks data must be processed by the preprocessing module. The preprocessing module will bring receiving data to the knowledge base format and use dictionary for conversation vectors to words and back.
 V.V. Petrov, N.V. Pavlova. Multi-Method Organization in Hybrid Expert Systems. Doklady Akademii Nauk. 1996, v. 350, n4. p. 465-466.
 A.V. Gavrilov. Architecture of the “Two-Hemisphere” Expert System. In the trans. of scient. works “Artificial Intelligence Systems”. 1993, p. 10-14. (on russian)
 C. S. Herrmann. A Hybrid Fuzzy-Neural Expert System for Diagnosis. In Proceeding of IJCAI, Montreal, Canada, August 1995.
 V. Wojcik. Hybrid Expert System (Rule Based/Neural Network) for Stock Portfolio Management.
 L. Zade. Notion of the Linguistic Variable and Its Using to Acceptance for the Approximate Decisions, Moscow: Mir, 1976.