NEURO-FUZZY CLASSIFIER OF STATE OF THE TECHNOLOGICAL PROCESS

Abstract

The algorithmic structure of the classifier of the state of the technological process is proposed, which provides processing of data coming through several channels of information support of the cyberphysical system. The structure contains an ensemble of deep recurrent neural networks and an output hybrid neural network. The ensemble solves the regression problems for information channels, and the output network serves for their generalization and classification based on fuzzy logic methods. The proposed hybrid architecture makes it possible to take advantage of two methodologies for constructing neural networks – to perform a retrospective analysis of time series using the DRNN ensemble and to generalize the results of their work by the ANFIS system. The structure of the software developed for simulation experiments is described and their results are presented.