A Federated Learning Approach Based on Hypernetworks for Wind Power Prediction
In distributed wind power systems, wind power generation prediction is affected by the non-independent and identically distributed (non-IID) nature of data, leading to low accuracy and generalization capabilities of prediction models. This paper proposes a federated wind power prediction algorithm based on hypernetworks. By deploying a hypernetwork on the server side, personalized gated recurrent unit (GRU) models are generated according to the client descriptors transmitted by each client for power prediction. Experiments conducted with real wind power operation data demonstrate that this method effectively enhances model performance while reducing training costs, exhibiting good application value.
Authors
- Yuange Liu (ORCID: https://orcid.org/0000-0001-6044-5392)
- Weishan Zhang (ORCID: https://orcid.org/0000-0001-9800-1068)
- Wenke Zang (ORCID: https://orcid.org/0000-0002-4899-6316)
- Hailin Feng (ORCID: https://orcid.org/0000-0003-2734-480X)
- Shaohua Cao (ORCID: https://orcid.org/0000-0001-8287-2942)
- Xiuxian Zhang (ORCID: https://orcid.org/0000-0002-2267-1716)
- Kai Fang (ORCID: https://orcid.org/0000-0003-0419-1468)
- Thippa Reddy (ORCID: https://orcid.org/0009-0000-2679-6404)
- Mohammad Kamrul Hasan (ORCID: https://orcid.org/0000-0001-5511-0205)
- Ruisheng Zhang (ORCID: https://orcid.org/0000-0002-6585-2656)
Institutions
- Zhejiang A & F University (CN)
- China University of Petroleum, East China (CN)
- National University of Malaysia (MY)
Publication Details
- Published
- 2024-12-13
- DOI
- https://doi.org/10.1109/icftic64248.2024.10912988
- Citations
- 1
- Primary Topic
- Energy Load and Power Forecasting
- Type
- article
- Field-Weighted Citation Impact
- 0.21
Funders
- National Natural Science Foundation of China
- Natural Science Foundation of Shandong Province