Hybrid neural network models for monitoring time series data of complex objects

Abstract

The problem of monitoring the state of complex objects of various natures based on classification and regression analysis of time series data is considered. Hybrid neural network models of classification and regression analysis are developed and studied using data on the functioning of three types of systems: spacecraft, information system and economic system, presented in the form of time series. For all types of systems, the proposed hybrid models demonstrate an advantage in accuracy. A genetic algorithm is developed for the automatic search of hybrid neural network models, with the help of which models of varying complexity are generated with an accuracy no lower than for models developed manually. As a result of the search, it is noted that the generated hybrid neural networks show results close to the maximum value of the fitness function. The fact is considered as experimental confirmation of the constructed solution to be close to optimal for certain search parameters.