10. Forecasting wind capacity using gauss regression models based on nuclear functions

Prognozowanie mocy wiatrowej przy wykorzystaniu modeli regresji procesu gaussa opartych na funkcjach jądra
 Maiusz Niekurzak
wind energy forecasting, machine learning, renewable energy
Zamieszczony w: 
Energy obtained from wind is characterized by high variability of output power and low availability, therefore it is necessary to predict their future energy production for economic purposes. The aim of the work is to develop and investigate forecasting methods allowing to increase the accuracy of wind farm power prediction with the use of artificial intelligence methods. The study explores the potential of various Gaussian Process Regression (GPR) models based on nuclear functions. In particular, the proposed models are dedicated to short-term prediction. The analysis was based on the operation of 6 wind farms located in Poland. A total of 120 sets of input data were selected for training and testing of the proposed models, and then their impact on the change in the accuracy of the forecasts was verified. As a result of the research, various structures of prognostic models were proposed and tested in order to select the most favorable variant. The comparison of the developed models based on the GPR method shows that the Pearson Universal Kernel (PUK) model was the superlative model. The prediction results of the GPR-PUK model turned out to be the most accurate and in line with the actual values, which allowed to verify the feasibility and effectiveness of using this model for forecasting. Machine learning methods will be able to show higher efficiency with the increase in the amount of data and the expansion of the set of potential explanatory variables. In the sea of data, machine learning methods are able to create predictive models more effectively without the need for tedious analyst interference in data preparation and multi-stage analysis. They will also allow for any frequent updating of the form of forecasting models, even after each addition to the data set.