DESCRIPTION OF AUTOSSA SOFTWARE FOR WINDOWS WITH GUI (c) Theodore Alexandrov, autossa@gmail.com Last modified: 17 Oct 2007 Introduction ----------------------------------------------------------------------- AutoSSA software for Windows with GUI is made of four modules: * Trend extraction; * Periodics extraction; * Signal extraction; * Signal forecasting. This software allows to load time series data from a file (text or Excel file) and automatically perform one of the tasks mentioned above. It is managed by setting thresholds and parameters for identification methods. AutoSSA provides information about eigen triples and output intermediate results of identification methods. Trend extraction ----------------------------------------------------------------------- This module contains the following trend identification methods: * Kendall; * Roots; * LowFreq (use this method by default). Periodic components extraction ----------------------------------------------------------------------- This module aims to: * extract all amplitude-modulated sine waves; * extract all periodic components; * extract periodic components with given period. Signal extraction ----------------------------------------------------------------------- We assume that a time series consists of trend, periodic components and residual, generally considered as a noise. The first two components constitute a signal of time series. Signal extraction is performed by extracting trend and periodic components. This module allows to solve collision situations, when a SVD component identified as corresponding both to a trend and a periodic component. Results are represented in a table, which helps to group SVD components. Signal forecasting ----------------------------------------------------------------------- Once a signal is extracted, it can be forecasted. The forecast in the framework of Caterpillar-SSA is performed in several steps: 1. identification of SVD components which correspond to a signal; 2. signal reconstruction; 3. finding the linear recurrent formula (LRF) corresponding to the signal; 4. prolongation of the signal using this LRF, where the LRF is applied to the last points of the signal.