Forecasting wind power generation with ensembling techniques

Date

2024-9

Author

Özertaş, Yunus Emre

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

40
views

160
downloads

Cite This

The global energy demand need is increasing with the growth in population and industrial developments. The increasing demand necessitates a transition to renewable energy sources, which are restored at a rate exceeding their consumption. Forecasting the future production of wind turbines is critical for strategic planning within the renewable energy sector. Various approaches, including statistical modeling, machine learning and deep learning techniques, have been proposed for this purpose, each varying interpretability and accuracy. This study aims to enhance the accuracy and interpretability of wind turbine output forecasts by employing ensemble techniques that integrate multiple modeling approaches. Specifically, data from wind turbines in Akhisar, Türkiye were used to predict short-term (1 hour ahead) and mid-term (1 day ahead). The motivation for combining different models arises from the lack of a fully comprehensive time series structure in the data and the absence of a universally optimal model for all time steps. Our combination method is based on predicting which model outperforms other ones in future timesteps. To identify the optimal model among various approaches, we proposed different combination techniques. As a result, the classifier that identifies the optimal model for future timesteps increased forecast accuracy. Additionally, adding errors of base models from previous timesteps to feature set of classifiers provided more accurate forecasts.

Subject Keywords

Ensemble architecture, Machine learning, Statistical learning, Time series, Wind power forecasting

URI

https://hdl.handle.net/11511/111443

Collections

Graduate School of Natural and Applied Sciences, Thesis