Fedorov E. Methods creation computer systems processor for processing natural language objects and signals

The dissertation is devoted the decision of a problem of insufficient efficiency of diagnostics of a current state of electromechanical object (EMO) by means of computer system (CS) and natural language interaction of the operator and computer system (CS). For the decision of this problem it is necessary to create and improve within the limits of the uniform approach methods of diagnostics and methods of natural language interaction for highly effective processors CS which are based on noise characteristics of EMO, vocal features of the operator, grammatical rules of transformation and interrelation of natural language objects (NLO). Are improved: methods of definition of parameters of a noise signal which are based on the spectral analysis, a threshold estimation and informative frequencies that allows to diagnose a state of technical object in due time; methods of identification of noise signals which lean against models neural network MLNN, trained by means of genetic algorithm with imitation annealing and using as reference pattern data of acoustic passports that allows to use an adaptive normalized threshold for estimation result of diagnostics of a state of objects and the probability of diagnostics increases.