Modelization of intermittent energies using non-stationary time series

Details

Supervisors

Van Bellegem, Sébastien

Faculty

Faculté des sciences

Degree label

Master [120] en science des données, orientation statistique, à finalité spécialisée

Abstract

Renewable energy sources could be a large part of the solution to the global warming problem. Current research is focused on increasing the use of intermittent renewable energies (wind and solar) in the power grid. The intermittent nature of these energy sources requires predictive research for their management in electrical networks. In this dissertation, we are particularly interested in the modelization and prediction of wind energy production in Belgium using a wavelet network algorithm on non-stationary time series. The interest of using the wavelet transform is to work simultaneously in two spaces, time and frequency, a considerable advantage over the sliding window Fourier transform. The wavelets move on temporal space of the signal by being dilated or contracted. The wavelet transform then measures the correlation between the signal and the wavelets, allowing finer resolutions to be achieved. This feature of the wavelet transform decomposes the signal according to a multiresolution analysis, which means that the signal is decomposed into different components. This multiresolution uses filters that transition from a coarse scale to increasingly finer scales. The prediction methodology employs an artificial neural network which uses as input the wavelet transform of the signal, using the "WaveletANN" package available on R.