Hybrid Speculation Model of Energy Consumption Based on Multivariate Singular Spectrum Analysis and Neural Fuzzy Network

Abstract

The results of the investigation of a hybrid model for short-term speculation of daily active energy consumption graphs in Moscow based on the Multivariate Singular Spectrum Analysis method (MSSA) and the neural fuzzy network (NFN) are presented taking into account factual and speculative data of air temperature and daylight illumination. In the MSSA module of the hybrid model the time series of energy consumption and meteorological factors are decomposed into independent components such as additive, trend, harmonic, and random components used in the NFN module are formed. While forming the speculative daily graph of energy consumption on the following day we used the archival data of active energy consumption diagrams and meteorological factors for 30 days (15 days preceding the speculated day in the current year and 15 days after the speculated date from the preceding year). For analyzing and transforming the time series of factual and speculative data of meteofactors we took data from the meteorological hardware-software complex (HSC “Meteo”). The results of the analysis of the speculation quality of daily active energy consumption graphs for the month from February 1 to February 30, 2016 are presented. The results of the speculations obtained from the hybrid model were estimated from the expectation values of the average daily ratio error (MAPE) and compared with the speculation errors obtained by the NFN model of the tentative decomposition of the source data time series into additive components.