Improved clustering and deep learning based short-term wind energy forecasting in large-scale wind farms

As a promising renewable solution for sustainable power generation worldwide, wind energy is receiving continuing attention from both industry and the academic community. However, the randomness and intermittency of wind energy will affect the stable operation and stability of the power system and f...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of renewable and sustainable energy 2020-11, Vol.12 (6)
Hauptverfasser: Huang, Yu, Li, Jiayu, Hou, Weizhen, Zhang, Bingzhe, Zhang, Yan, Li, Yongling, Sun, Li
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:As a promising renewable solution for sustainable power generation worldwide, wind energy is receiving continuing attention from both industry and the academic community. However, the randomness and intermittency of wind energy will affect the stable operation and stability of the power system and further affect the economic benefits of the power grid. What makes the matter worse is the inevitable coupling between each pair of wind turbines in the large-scale wind farm. Besides, the resolution of prediction is severely limited by the spatial scale of wind farms. These problems bring great difficulties for the control and scheduling of wind farms. To this end, this paper proposes a novel wind speed prediction method for wind farms by borrowing some wisdom from machine learning methods. First, density peak clustering (DPC) is employed to separate the tremendous number of scattered wind turbines into a much significantly reduced number of groups, the wind turbines in each of which are treated as a unity. Based on the priority setting of each indicator in clustering, the data are preprocessed with different weightings. Principal component analysis is utilized to avoid DPC's poor clustering effects in case the dataset is high-dimensional. Finally, by considering simultaneous effects from historical and present data, long short-term memory based deep learning neural networks are trained and used to iteratively predict the potential of the wind energy in each unit for each time slot. The effectiveness of the proposed algorithm is verified by taking an in-service wind farm in China as an example.
ISSN:1941-7012
1941-7012
DOI:10.1063/5.0016226