CASPIAN JOURNAL

MANAGEMENT AND HIGH TECHNOLOGIES

Neuroevolutional method of decision making intellectualization in conditions of indeterminacy

Read

Khlopkova Olga A. Neuroevolutional method of decision making intellectualization in conditions of indeterminacy // Caspian journal : management and high technologies. — 2015. — №3. — pp. 114-129.

Khlopkova Olga A. - post-graduate student, Moscow State University of Economics, Statistics and Informatics, 7 Nezhinskaya St., Moscow, 119501, Russian Federation, ohlopkova@nifi.ru

This article presents the adaptive method of intellectualization of decision support systems under conditions of indeterminacy, incomplete information and dynamic environment. As distinct from existing neuroevolutional methods such as GNARL, ENS3, NEAT (and its modifications including HyperNEAT), the method is applicable for the optimization of both the structure and parameters of the neural networks of any topology. It is deprived of the main neuroevulution-related drawbacks such as competing conventions, unprotected innovations, initialization and topology minimization, topological innovations. This method reduces the risk of premature convergence, network paralysis and solves the local minima problem. This article describes the main aspects of the method: the scheme structure defined by the direct encoding, chromosomes parameters, fitness function definition and algorithms of selection, mutations and recombination. The article introduces the theoretical rationale for the memetic phase of the parameters and structure local tuning. The article gives examples of successful application of the method to classical neuroevolutional problems such as adaptive control, constructing of logical functions, damaged data recovery. The method proved its efficiency in appliance to practical problems of information system protection from DDoS-attacks, rating calculation of materials in knowledge base and improvement of the 3D-printing systems.

Key words: генетические алгоритмы, нейронные сети, искусственный интеллект, нейроэволюция, меметичные алгоритмы, системы поддержки принятия решений, неопределенность, эволюционные вычисления, гибридные интеллектуальные системы, TWEANN, COGANN, 3D-принтеры, genetic a