?
Сегментация существенно нестационарных временных рядов методом нелинейного факторного анализа
С. 287-297.
Perminov G. I.
Abstract
In this paper we consider essentially non-stationary time series, ie in which the change in the structure, variables, or model coefficients of the variables. For such series found the methods of reduction to stationarity, so the analysis and prediction of a number of proposed segmentation - dividing it into segments with a quasi-stationary characteristics of statistical distributions. The paper for this purpose, consider the use of the method of nonlinear factor analysis (PFA) in comparison with linear FA, the method of moving reference models and R / S analysis on the discrete wavelet decomposition. Analysis of results showed the applicability of the essentially non-stationary series of all four methods. Of these, the most accurate results with low labor costs is the method of the APF.
In book
Perminov G. I. Севастополь : ТНУ им. В.И. Вернадского, 2012