Authors: A.E. Kreneva, A.I. Kolokolnikova

Title of the article: Applying singular spectrum analysis in forecasting indicators of energy consumption

Year: 2022, Issue: 3, Pages: 50-58

Branch of knowledge: 09.05.03 Electrotechnical complexes and systems

Index UDK: 004.942

DOI: 10.26730/1816-4528-2022-3-50-58

Abstract: Based on the well-known method of singular spectral analysis of time series (SSA) and its modifications (МSSA и ASSA), an algorithm and a program for long-term forecasting of energy consumption indicators have been developed. In the proposed forecasting algorithm αSSA, the use of the ideas of regularization and weighting of the series expansion components allows us to obtain a range of forecasting models. In this paper, specific ways of forming alternative models are implemented in the form of an algorithm and a program. To evaluate alternative models, a quality criterion has been formulated, in which mismatches between the predictive and actual values of the series are used to assess the effectiveness of the model. For long-term forecasting purposes, the quality criterion includes assessing the accuracy of long-term forecasts by the depth of the forecast specified as an algorithm parameter. The quality criterion is calculated on the basis of long-term forecasts of the last elements of a given number series. Another feature of the presented algorithm is the accounting of the "seasonality" of the series, expressed in some similarity of daily hourly charts. Forecasting for a certain hour of the day is made on the basis of data for a given hour of the day, extracted from the hourly schedule. Used technique allows to speed up the work of the program, where the operation of finding eigenvectors and eigenvalues of the matrix is critical. The dimension of the task in this case is reduced by 24 times. The proposed algorithm was implemented by means of the object-oriented programming language Delphi. A numerical study was conducted on energy consumption data in the Siberian price zone. These examples of forecasting prove the effectiveness of the developed algorithm and the created application.

Key words: time series long-term forecasting singular expansion singular spectrum analysis «Caterpillar»-SSA noise

Receiving date: 13.05.2022

Approval date: 20.06.2022

Publication date: 01.07.2022

CITE DOWNLOAD