DECOMPOSIÇÃO DE SÉRIES TEMPORAIS UTILIZANDO OS ALGORITMOS SSA e SSA RECURSIVO

Abstract

In order to extract temporal characteristics, find the trend and seasonality and eliminate noise from a time series, one of the ways is to use a time series decomposition technique. In this work, two time series decomposition methods are discussed. The first performs the SSA algorithm for time series decomposition using a more usual approach, in which eigenvalues and eigenvectors are calculated in batch from the whole time series. In the second approach, the eigenvalues and eigenvectors are updated recursively. The methods were implemented in MatLab® and the computational cost between the algorithms was compared. The results illustrate that the recursive SSA approach has a lower computational cost compared to the batch method.