EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Short-term motion prediction of floating offshore wind turbine using inverted Transformer deep learning frameworks

Xinheng Zhang, Yao Cheng, Puyi Yang, Renwei Ji, Pei Zhang, Xiaosen Xu, Shuaishuai Wang

Clean Energy · 09.07.2026 · 10 (4) · 68-84

DOI: 10.1093/ce/zkag025 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Clean Energy, online erschienen am 02.06.2026. Verfasst von Xinheng Zhang und 6 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Jiangsu Province, Shanghai Jiao Tong University.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Floating offshore wind turbine (FOWT) platforms are critical components of offshore wind development. Accurately predicting their motion response is vital for ensuring platform safety, optimizing power generation efficiency, and reducing operation and maintenance costs. Therefore, drawing on advancements in deep learning, this study proposes a method utilizing the inverted Transformer (iTransformer) model to predict the short-term motion response of FOWTs. The primary objective is to enhance prediction accuracy. First, numerical simulations of a specific 5 MW FOWT platform were conducted under various environmental conditions. Subsequently, the simulation data were preprocessed and utilized for experimental evaluation. Finally, to assess model performance, comparative analyses were conducted between the iTransformer model and mainstream recurrent neural network (RNN) hybrid models currently used in this field. Specifically, long short-term memory (LSTM)-attention, gated recurrent units (GRU)-attention, and bidirectional LSTM-attention models were compared under different predicted advance times (PATs). The results indicate that at a PAT of 3 s, the iTransformer model achieves the highest accuracy, although the difference compared to the three RNN hybrid models is not significant. However, as the PAT increases, the superiority of the iTransformer model becomes increasingly apparent. Specifically, under the EC1 condition at a PAT of 9 s, compared with the best-performing RNN-attention baseline among the LSTM-attention model, GRU-attention model, and bidirectional LSTM-attention model, the iTransformer reduces the mean absolute error values of sway, roll, and pitch by 47.4%, 44.6%, and 43.2%, respectively. Furthermore, the model demonstrates a significant improvement in training speed. These findings suggest that the proposed model outperforms other models in terms of both prediction accuracy and training efficiency. Additionally, it offers an extended forecasting horizon. This study validates the potential of the iTransformer as a reliable tool for short-term motion prediction, providing robust support for planning the safe operation and maintenance of FOWTs.

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Jiangsu Province
  • Shanghai Jiao Tong University
  • Open Fund of State Key Laboratory of Coastal and Offshore Engineering

Zitieren

APA

Zhang, X., Cheng, Y., Yang, P., Ji, R., Zhang, P., Xu, X., & Wang, S. (2026). Short-term motion prediction of floating offshore wind turbine using inverted Transformer deep learning frameworks. Clean Energy, 10(4), 68-84. https://doi.org/10.1093/ce/zkag025

Vancouver

Zhang X, Cheng Y, Yang P, Ji R, Zhang P, Xu X, et al. Short-term motion prediction of floating offshore wind turbine using inverted Transformer deep learning frameworks. Clean Energy. 2026;10(4):68-84. doi:10.1093/ce/zkag025

BibTeX
@article{Zhang2026,
  author = {Zhang, Xinheng and Cheng, Yao and Yang, Puyi and Ji, Renwei and Zhang, Pei and Xu, Xiaosen and Wang, Shuaishuai},
  title = {Short-term motion prediction of floating offshore wind turbine using inverted Transformer deep learning frameworks},
  journal = {Clean Energy},
  year = {2026},
  volume = {10},
  number = {4},
  pages = {68--84},
  doi = {10.1093/ce/zkag025}
}
Kostenlos starten →