Lamonova N. Detection of properties changes in stochastic sequences using artificial neural networks.

This dissertation is devoted to the development of methods for detection of properties changes in stochastic sequences for the tasks of early real time diagnostics of nonlinear stochastic plants using artificial neural networks (ANN). The developed algorithms have finite memory based on the sliding window and exponential weighting and allow to solve the problem of changes detection in the plant properties in real time. In order to improve properties of the algorithms, the rules of the memory level correction have been proposed. It has been shown that utilization of the mutlilayer perceptron in the tasks of early diagnostics is a difficult problem. A software package for modeling the designed specialized ANN has been developed.