Time series Analysis

Abstract

This thesis is focused on the analysis of time series in the environment of energy markets and forecasting future energy consumption. In this area, there is no uniform methodology on how to approach this task. The lack of publicly available datasets complicates research. Preprocessing and postprocessing techniques are also not adequately covered in these areas. These are the main goals that this thesis focused on solving and bringing progress in those areas. Therefore, this work can be beneficial for anyone interested in this field of research. The theoretical section describes aspects of time series forecasting in its general form. This description is extended by the specialization of these techniques to the areas of natural gas and electricity consumption forecasting. We have prepared our own datasets and made part of them available to other researchers to use in their experiments. The designed methodology focused on the application of forecasting models on prepared datasets. A series of experiments validated the usability of this methodology and prepared datasets. Experiments were focused not only on machine learning algorithms but were also compared with classical statistical models. An important area was feature engineering. We have managed to further improve the precision of forecasts with preprocessing and postprocessing techniques. The benefit of those methods was measured and evaluated.