INFORMATION-EXTREME ALGORITHM OF UNIMODAL CLASSIFIER FOR GROWING SINGLE CRYSTAL DIAMETER EVALUATION

A.S. Dovbish, V.V. Moskalenko

This paper proposes information-extreme algorithm of analis and synthesis of learning and self-learning decision support system (DSS). It based on unimodal classifier characterized by single center of pattern vector-realizations distribution and structured alphabet of recognition classes. Training the DSS optimizes coltrol permits vector for identification signs by the simultaneous changing the parameter of coltrol permits field for all the signs.

As a criterion of the functional efficiency of training was considered modified normalized Kullback information criterion. Training parameters optimization carries iterative search for the global maximum Kullback criterion in the allowable domain of it function definition. On the example of DSS for control of growing scintillate single crystals, the article considers implementation the algorithm. According results of physical modeling unimodal classifier shown increase DSS training functional efficiency compared with multimodal classifier.

Keywords: optimization, training, unimodal classifier, decision support system, management, scintillate single crystal.