[Dissertation] Improving forecast accuracy of wind power output using multi-input LSTM model.
Masters dissertation by Phillemon, African Institute for Mathematical Sciences
Members
Phillemon Ntona Senoamadi, AIMS Big Data Science
Supervisor(s)
- Dr. Nicolene Botha, CSIR
- Dr. Bubacarr Bah, AIMS
- Dr. Vukosi Marivate, University of Pretoria
Abstract
Accurate wind power forecasting can help operators to estimate how much profit to make depending on where and when in order to produce more electricity. Precise prediction allows operators to achieve suitable trading performances. In this study, data sets obtained from two wind stations i.e. Memel (Free State) and Jozini (Kwa-Zulu-natal), are used for the prediction of the mean wind speed. Moreover, the mean wind speed from one of the station is used to increase the prediction of the other station. The long short term memory (LSTM) model is used for the prediction which is benchmarked by the persistence forecast. The LSTM model was able to beat the benchmark model root mean square results. The applied forecasting models focuses on the first, twelfth and twenty fourth hour.