An Adaptive Extrapolator of Nonstationary Random Processes

Abstract: Prediction of random processes is important in the development of systems of predictive diagnostics of the state of objects of electrical engineering, planning of load changes of generating power plants, and construction of regulators as a part of an operator that converts mismatch into control signals. This article considers an extrapolator the parameters of which change in time in order to minimize the prediction error of nonstationary random processes. The extrapolator uses Chebyshev polynomials that are orthogonal to a set of equidistant points with the coefficients of the predictive polynomial obtained by the least-squares method. The number of points on the time axis in which the value of the random process is known before the prediction interval, as well as the degree of the predicting polynomial, change in time depending on a forecast error that took place earlier. The work of the extrapolator is formalized, the algorithm of its functioning is described, and various rules of control of the mechanism of parameter change are considered. The results of simulation experiments testify to the efficiency of using the adaptive extrapolator for prediction of nonstationary random processes. © 2023, Allerton Press, Inc.