Heinermann P.J. Wind power prediction with machine learning ensembles

University of Oldenburg, 2016. — 170 p.For a sustainable integration of wind power into the electricity grid, precise and robust predictions are required. Machine learning methods can be used as purely data-driven, spatio-temporal prediction models that yield better results than traditional physical models based on weather simulations. The objectives of this thesis are the improvement of prediction errors and the reduction of computation times. This thesis proposes a robust and practical prediction framework based on ensembles, which combines the predictions of numerous and preferably diverse models. A comprehensive experimental evaluation shows that the combination of different techniques to an ensemble outperforms state-of-the-art prediction models while requiring shorter computation times. For model selection we employ evolutionary multi-objective optimization algorithms. The methods offer an efficient and comfortable balancing of preferably low prediction errors and moderate computational costs.