Method for forecasting changes in time series parameters in digital information management systems

Abstract

Predicting changes in the parameters of time series is of high significance when monitoring the research processes in digital information management systems. This task also arises when researching the issues of increasing the prediction horizon and minimizing the forecast error. In this paper, we investigate prediction algorithms based on models that reproduce the dynamics of a time series in the form of artificial neural networks. We also consider the development of algorithms for control, functioning and training of an artificial neural network in a matrix form and obtaining an algorithm for the return substitution, with the help of which it is possible to obtain an increase in the depth of the forecast. The paper presents the solution of the prediction problem consisting in finding prediction estimates by minimizing the loss function - the square of the norm of estimate deviation from the observed values of the time series and in determining the model coefficients by using an artificial neural networks learning algorithm based on the iterative method of back-propagating errors. Application of the developed algorithms has allowed us to build a structural scheme for implementing neural network forecasting, with the help of which it is possible to obtain a fairly accurate representation of changes in the parameters of time series in the process monitoring systems in terms of the runtime and the minimized error of the forecasting.