Methods for optimization of identification of nonstationary objects on the basis of neurofuzzy network with adjustment of computing circuits parameters.

The problem is formulated and methodical bases are developed for constructing the technology of data intellectual analysis for recognition, classification and forecasting of non-stationary objects. Methods of use the generalized properties and features of fuzzy identification models, segmentation of casual time series, updating of computing circuits of neurofuzzy networks (NFN), adaptive training of NFN, adjustment of parameters of structural components are offered to increase the quality of functioning for systems to monitoring industrial-technological complexes. Algorithms are developed to choice the suitable membership function (MF) for input and output linguistic variable, regulation of MF borders, interval of values of fuzzy sets carrier by compression and expansion. The rules are developed to define the effective modes of NFN training with optimum number of iterations to ensure identification of poorly formalizable processes in conditions of insufficiency of a priory information and large parametrical uncertainty with large accuracy. Results of researches are realized as a program complex to fuzzy identification and data processing of non-stationary objects.